Book Catalyst Design Optimal Distribution of Catalyst in Pellets Reactors and Membranes

Catalyst Design
Heterogeneous catalysis is widely used in chemical, refinery, and pollution-

control processes. For this reason, achieving optimal performance of cat-
alysts is a significant issue for chemical engineers and chemists. This book
addresses the question of how catalytic material should be distributed
inside a porous support in order to obtain optimal performance. It treats
single- and multiple-reaction systems, isothermal and nonisothermal con-
ditions, pellets, monoliths, fixed-bed reactors, and membrane reactors.
The effects of physicochemical and operating parameters are analyzed to
gain insight into the underlying phenomena governing the performance
of optimally designed catalysts. Throughout, the authors offer a balanced
treatment of theory and experiment. Particular attention is given to prob-
lems of commercial importance. With its thorough treatment of the de-
sign, preparation, and utilization of supported catalysts, this book will
be a useful resource for graduate students, researchers, and practicing
engineers and chemists.
Massimo Morbidelli is Professor of Chemical Reaction Engineering in

the Laboratorium für Technische Chemie at ETH, Zürich.
Asterios Gavriilidis is Senior Lecturer in the Department of Chemical

Engineering at University College London.
Arvind Varma is the Arthur J. Schmitt Professor in the Department of

Chemical Engineering at the University of Notre Dame.
CAMBRIDGE SERIES IN CHEMICAL ENGINEERING
Series Editor:
Arvind Varma, University of Notre Dame
Editorial Board:
Alexis T. Bell, University of California, Berkeley
John Bridgwater, University of Cambridge
Robert A. Brown, MIT
L. Gary Leal, University of California, Santa Barbara
Massimo Morbidelli, ETH, Zurich
Stanley I. Sandler, University of Delaware
Michael L. Shuler, Cornell University
Arthur W. Westerberg, Carnegie Mellon University
Books in the Series:

E. L. Cussler, Diffusion: Mass Transfer in Fluid Systems, second edition
Liang-Shih Fan and Chao Zhu, Principles of Gas-Solid Flows
Hasan Orbey and Stanley I. Sandler, Modeling Vapor-Liquid Equilibria:
Cubic Equations of State and Their Mixing Rules
T. Michael Duncan and Jeffrey A. Reimer, Chemical Engineering Design
and Analysis: An Introduction
A. Varma, M. Morbidelli and H. Wu, Parametric Sensitivity in Chemical
Systems
John C. Slattery, Advanced Transport Phenomena
M. Morbidelli, A. Gavriilidis and A. Varma, Catalyst Design: Optimal
Distribution of Catalyst in Pellets, Reactors, and Membranes
Catalyst Design
OPTIMAL DISTRIBUTION OF CATALYST IN
PELLETS, REACTORS, AND MEMBRANES
Massimo Morbidelli
ETH, Zurich
Asterios Gavriilidis
University College London
Arvind Varma
University of Notre Dame
To our teachers and students
CAMBRIDGE UNIVERSITY PRESS

Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo
Cambridge University Press

The Edinburgh Building, Cambridge CB2 2RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521660594
© Cambridge University Press 2001
This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without
the written permission of Cambridge University Press.
First published 2001

This digitally printed first paperback version 2005
A catalogue record for this publication is available from the British Library
Library of Congress Cataloguing in Publication data

Morbidelli, Massimo.
Catalyst design : optimal distribution of catalyst in pellets, reactors, and membranes /
Massimo Morbidelli, Asterios Gavriilidis, Arvind Varma.
p. cm. – (Cambridge series in chemical engineering)
Includes bibliographical references.
ISBN 0-521-66059-9
1. Catalysts. I. Gavriilidis, Asterios, 1965– II. Varma, Arvind. III. Title. IV. Series.
TP159.C3 M62 2001
660´.2995 – dc21 00-041460
ISBN-13 978-0-521-66059-4 hardback

ISBN-10 0-521-66059-9 hardback
ISBN-13 978-0-521-01985-9 paperback

ISBN-10 0-521-01985-0 paperback
Contents
Preface page xiii
1 Introduction 1
1.1 Importance of Catalysis 1
1.2 Nonuniform Catalyst Distributions 1
1.3 Overview of Book Contents 4
2 Optimization of the Catalyst Distribution in a Single Pellet 6
2.1 The Case of a Single Reaction 6
2.1.1 Isothermal Conditions 6
2.1.2 Nonisothermal Conditions 15
2.1.3 Arbitrary Kinetics with External Transport Resistances 18
2.1.4 Dynamic Behavior 23
2.2 Multiple Reactions 25
2.3 The General Case of a Complex Reaction System 30
2.3.1 An Illustrative Example 31
2.3.2 General Reaction System 35
2.4 Catalyst Dispersion Considerations 40
2.4.1 Factors Affecting Catalyst Dispersion 40
2.4.2 Dependence of Catalytic Surface Area on Catalyst Loading 43
2.5 Optimal Distribution of Catalyst Loading 46
2.5.1 The Problem Formulation 47
2.5.2 A Single First-Order Isothermal Reaction 51
2.5.3 Linear Dependence between the Active Element Surface
Area and Its Loading 54
2.5.4 First-Order Nonisothermal Reactions: Numerical Optimization 55
2.5.5 Multistep Optimal Loading Distribution 59
2.6 Experimental Studies 63
2.6.1 Oxidation Reactions 63
2.6.2 Hydrogenation Reactions 66
2.6.3 Fischer–Tropsch Synthesis 68
vii
viii Contents
3 Optimization of the Catalyst Distribution in a Reactor 69

3.1 A Single Reaction 69
3.3.1 Propane and CO Oxidation 83
3.3.2 Catalytic Incineration of Volatile Organic Compounds 85
4 Studies Involving Catalyst Deactivation 86
4.1 Nonselective Poisoning 86
4.2 Selective Poisoning 89
4.3.1 Methanation 91
4.3.2 Hydrogenation 92
4.3.3 NO Reduction 94
5 Membrane Reactors 95
5.1 Membrane Reactors with Nonuniform Catalyst Distribution 95
5.2 Optimal Catalyst Distribution in Pellets for an
Inert Membrane Reactor 100
5.3 Optimal Catalyst Distribution in a Catalytic Membrane Reactor 100
5.4.1 Dehydrogenation Reactions 102
5.4.2 Preparation of Catalytic Membranes 105
6 Special Topics of Commercial Importance 110
6.1 Automotive Exhaust Catalysts 110
6.1.1 Design of Layered Catalysts 111
6.1.2 Nonuniform Axial Catalyst Distribution 113
6.2 Hydrotreating Catalysts 115
6.3 Composite Zeolite Catalysts 119
6.4 Immobilized Biocatalysts 121
6.5 Functionalized Polymer Resins 124
6.5.1 Preparation of Nonuniformly Functionalized Resin Particles 124
6.5.2 Applications to Reacting Systems 126
7 Preparation of Pellets with Nonuniform Distribution of Catalyst 131
7.1 Adsorption on Powders 132
7.1.1 Adsorption Isotherm Models 132
7.1.2 Effect of Impregnation Variables on Adsorption 135
7.1.2.a Solution pH and Nature of Support 135
7.1.2.b Surface Heterogeneity 138
7.1.2.c Ionic Strength 140
7.1.2.d Precursor Speciation 140
7.1.2.e Coimpregnants 142
7.1.2.f Nature of the Solvent 144
7.1.3 Surface Ionization Models 144
7.1.3.a Constant-Capacitance Model 145
Contents ix
7.1.3.b Diffuse-Layer Model 146

7.1.3.c Basic Stern Model 147
7.1.3.d Triple-Layer Model 147
7.1.3.e Four-Layer model 149
7.2 Simultaneous Diffusion and Adsorption in Pellets 149
7.2.1 Theoretical Studies 150
7.2.1.a Dry Impregnation 150
7.2.1.b Wet Impregnation 153
7.2.1.c Effects of Electrokinetic and Ionic Dissociation Phenomena 156
7.2.1.d Effect of Drying Conditions 156
7.2.2 Experimental Studies 158
7.2.2.a Single-Component Impregnation 159
7.2.2.b Multicomponent Impregnation 161
7.2.2.c Effects of Drying 165
7.2.2.d Determination of Catalyst Distribution 169
7.2.3 Comparison of Model Calculations with Experimental Studies 169
7.2.3.a Dry Impregnation 169
7.2.3.b Wet Impregnation 171
Appendix A: Application of the Maximum Principle for Optimization

of a Catalyst Distribution 181
Appendix B: Optimal Catalyst Distribution in Pellets for an Inert
Membrane Reactor: Problem Formulation 188
B.1 The Mass and Energy Balance Equations 188
B.2 The Performance Indexes 191
B.3 Development of the Hamiltonian 192
Notation 195
References 201
Author Index 221
Subject Index 225
Preface
Heterogeneous catalysis is used widely in chemical, refinery and pollution-control

processes. Current worldwide catalyst usage is about 10 billion dollars annually,
with ca. 3% annual growth rates. While these numbers are impressive, the eco-
nomic importance of catalysis is far greater since about $200–$1,000 worth of
products are manufactured for every $1 worth of catalyst consumed. Further, a
vast majority of pollution-control devices, such as catalytic converters for automo-
biles, are based on catalysis. Thus, heterogeneous catalysis is critically important
for the economic and environmental welfare of society.
In most applications, the catalyst is deposited on a high surface area support
of pellet or monolith form. The reactants diffuse from the bulk fluid, within the
porous network of the support, react at the active catalytic site, and the products
diffuse out. The transport resistance of the porous support alters the concen-
trations of chemical species at the catalyst site, as compared to the bulk fluid.
Similarly, owing to heat effects of reaction, temperature gradients also develop
between the bulk fluid and the catalyst. The consequence of these concentration
and temperature gradients is that reactions occur at different rates, depending on
position of the catalyst site within the porous support. In this context, since the
catalytic material is often the most expensive component of the catalyst-support
structure, the question naturally arises as to how should it be distributed within
the support so that the catalyst performance is optimized? This book addresses
this question, both theoretically and experimentally, for supported catalysts used
in pellets, reactors and membranes.
In Chapter 2, optimization of catalyst distribution in a single pellet is considered,
under both isothermal and nonisothermal conditions. Both single and multiple re-
action systems following arbitrary kinetics are discussed. Chapter 3 deals with op-
timization of catalyst distribution in pellets comprising a fixed-bed reactor, while
systems involving catalyst deactivation are addressed in Chapter 4. In Chapter
5, the effect of catalyst distribution on the performance of inorganic membrane
reactors is presented, where the catalyst is located either in pellets packed inside
an inert tubular membrane or within the membrane itself. Issues related to cata-
lysts of significant commercial importance, including automotive, hydrotreating,
xi
xii Preface
composite zeolite, biological, and functionalized polymer resin types, are ad-
dressed in Chapter 6. The final Chapter 7 considers catalyst preparation by im-
pregnation techniques, where the effects of adsorption, diffusion and drying on
obtaining desired nonuniform catalyst distributions within supports are discussed.
This book should appeal to all those who are interested in design, preparation and
utilization of supported catalysts, including chemical and environmental engineers
and chemists. It should also provide a rich source of interesting mathematical prob-
lems for applied mathematicians. Finally, we hope that industrial practitioners will
find the concepts and results described in this book to be useful for their work.
This book can be used either as text for a senior-graduate level specialized
course, or as a supplementary text for existing courses in reaction engineering,
industrial chemistry or applied mathematics. It can also be used as a reference for
industrial applications.
We thank our departmental colleagues for maintaining an atmosphere con-
ducive to learning. We also thank our families for their encouragement and sup-
port, which made this writing possible.
Massimo Morbidelli
Asterios Gavriilidis
Arvind Varma
1
Introduction
1.1 Importance of Catalysis

A large fraction of chemical, refinery, and pollution-control processes involve
catalysis. Its importance can be demonstrated by referring to the catalyst market.
In 1993 the worldwide catalyst usage was $8.7 billion, comprising $3.1 billion for
chemicals, $3 billion for environmental applications, $1.8 billion for petroleum
refining, and $0.8 billion for industrial biocatalysts (Schilling, 1994; Thayer, 1994).
The total market for chemical catalysts is expected to grow by approximately
20% between 1997 and 2003, primarily through growths in environmental and
polymer applications (McCoy, 1999). For the U.S., the total catalyst demand was
$2.4 billion in 1995 and is expected to rise to $2.9 billion by the year 2000 (Shelley,
1997). While these figures are impressive, the economic importance of catalysis is
even greater when considered in terms of the volume and value of goods produced
through catalytic processes. Catalysis is critical in the production of 30 of the top
50 commodity chemicals produced in the U.S., and many of the remaining ones
are produced from chemical feedstocks based on catalytic processes. In broader
terms, nearly 90% of all U.S. chemical manufacturing processes involve catal-
ysis (Schilling, 1994). Although difficult to estimate, approximately $200–$1000
(Hegedus and Pereira, 1990; Cusumano, 1991) worth of products are manufac-
tured for every $1 worth of catalyst consumed. The value of U.S. goods produced
using catalytic processes is estimated to be between 17% and 30% of the U.S.
gross national product (Schilling, 1994). In addition, there is the societal benefit of
environmental protection, since emission control catalysts are a significant sector
of the market (McCoy, 1999).
1.2 Nonuniform Catalyst Distributions

The active materials used as catalysts are often expensive metals, and in order
to be utilized effectively, they are dispersed on large-surface-area supports. This
approach in many cases introduces intrapellet catalyst concentration gradients
during the preparation process, which were initially thought to be detrimental
1
2 Introduction
to catalyst performance. The effects of deliberate nonuniform distribution of the

catalytic material within the support started receiving attention in the 1960s.
Early publications which demonstrated the superiority of nonuniform catalysts
include those of Mars and Gorgels (1964), Michalko (1966a,b), and Kasaoka and
Sakata (1968). Mars and Gorgels (1964) showed that catalyst pellets with an inert
core can offer superior selectivity during selective hydrogenation of acetylene in
the presence of a large excess of ethylene. Michalko (1966a,b) used subsurface-
impregnated Pt/Al2 O3 catalyst pellets for automotive exhaust gas treatment and
found that they exhibited better long-term stability than surface-impregnated pel-
lets. Kasaoka and Sakata (1968) derived analytical expressions for the effective-
ness factor for an isothermal, first-order reaction with various catalyst activity
distributions and showed that those declining towards the slab center gave higher
effectiveness factors. A number of publications have dealt with analytical calcula-
tions of the effectiveness factor for a variety of catalyst activity distributions. These
include papers by Kehoe (1974), Nyström (1978), Ernst and Daugherty (1978),
Gottifredi et al. (1981), Lee (1981), Do and Bailey (1982), Do (1984), and Papa
and Shah (1992). Some researchers have focused on the issue of shape and activity
distribution normalization, where the objective is to provide generalized expres-
sions for the catalytic effectiveness (Wang and Varma, 1978; Yortsos and Tsotsis,
1981, 1982a,b; Morbidelli and Varma, 1983).
Pellets with larger catalyst activity in the interior than on the surface can result
in higher effectiveness factors in the case of reactions which behave as negative-
order at large reactant concentrations, such as those with bimolecular Langmuir–
Hinshelwood kinetics (Villadsen 1976; Becker and Wei, 1977a). Nonuniform cat-
alyst distributions can also improve catalyst performance for reactions following
complex kinetics (Juang and Weng, 1983; Johnson and Verykios, 1983, 1984). For
example, in multiple-reaction systems, catalyst activity distribution affects selec-
tivity. Shadman-Yazdi and Petersen (1972) and Corbett and Luss (1974) studied
an irreversible isothermal first-order consecutive reaction system for a variety of
activity profiles. Selectivity to the intermediate species was favored by distribu-
tions concentrated towards the external surface of the pellet. Juang and Weng
(1983) studied parallel and consecutive reaction systems under nonisothermal
conditions. Which catalyst profile amongst those considered gave the best selec-
tivity depended on the characteristics of the particular reaction system. Johnson
and Verykios (1983, 1984) and Hanika and Ehlova (1989) studied parallel reaction
networks and showed that nonuniform activity distributions can enhance selectiv-
ity. Similar improvements were also demonstrated by Cukierman et al. (1983) for
the van de Vusse reaction network. Ardiles et al. (1985) considered a bifunctional
reacting network representative of hydrocarbon reforming, and showed that se-
lectivity to intermediate products was influenced by the distribution of the two
catalytic functions.
The effects of nonuniform activity in catalyst pellets have also been studied
in the context of fixed-bed reactors. Minhas and Carberry (1969) studied numer-
ically the advantages of partially impregnated catalysts for SO2 oxidation in an
adiabatic fixed-bed reactor. Smith and Carberry (1975) investigated the produc-
tion of phthalic anhydride from naphthalene in a nonisothermal nonadiabatic
1.2 Nonuniform Catalyst Distributions 3
fixed-bed reactor. This is a parallel–consecutive reaction system for which the in-
termediate product yield is benefited by a pellet with an inert core. Verykios et al.
(1983) modeled ethylene epoxidation in a nonisothermal nonadiabatic fixed-bed
reactor with nonuniform catalysts. They showed that improved reactor stability
against runaway could be obtained, along with higher reactor selectivity and yield,
as compared to uniform catalysts.
Rutkin and Petersen (1979) and Ardiles (1986) studied the effect of activity
distributions for bifunctional catalysts in fixed-bed reactors, for the case of multiple
reaction schemes. Each reaction was assumed to require only one type of catalyst.
It was shown that catalyst activity distributions had a strong influence on reactant
conversion and product selectivities.
Nonuniform activity distribution for catalysts experiencing deactivation has
been studied by a number of investigators (DeLancey, 1973; Shadman-Yazdi and
Petersen, 1972; Corbett and Luss, 1974; Becker and Wei, 1977b; Juang and Weng,
1983; Hegedus and McCabe, 1984). If deactivation occurs by sintering, it is mini-
mized by decreasing the local catalyst concentration, i.e., a uniform catalyst offers
the best resistance to sintering (Komiyama and Muraki, 1990).
In all cases considered above, catalyst performance was assessed utilizing ap-
propriate indexes. The most common ones include effectiveness, selectivity, yield,
and lifetime. Effectiveness factor relates primarily to the reactant conversion that
can be achieved by a certain amount of catalyst, while selectivity and yield relate
to the production of the desired species in multiple reaction systems. In the case of
membrane reactors additional performance indexes (e.g. product purity) become
of interest. In deactivating systems, other indexes incorporating the deactivation
rate can be utilized apart from catalyst lifetime. Another index, which has not
been employed in optimization studies because it is difficult to express in quanti-
tative terms, is attrition. Catalyst pellets with an outer protective layer of support
are beneficial in applications where attrition due to abrasion or vibration occurs,
since only the inert and inexpensive support is worn off and the precious active
materials are retained.
The key parameters which control the effect of nonuniform distribution on the
above performance indexes are reaction kinetics, transport properties, operating
variables, deactivation mechanism, and catalyst cost. All the early studies dis-
cussed above demonstrated that nonuniform catalysts can offer superior conver-
sion, selectivity, durability, and thermal sensitivity characteristics to those wherein
the activity is uniform. This was done by comparing the performance of catalysts
with selected types of activity profiles, which led to the best profile within the
class considered, but not to the optimal one. Morbidelli et al. (1982) first showed
that under the constraint of a fixed total amount of active material, the optimal
catalyst distribution is an appropriately chosen Dirac-delta function; i.e., all the
active catalyst should be located at a specific position within the pellet. This dis-
tribution remains optimal even for the most general case of an arbitrary number
of reactions with arbitrary kinetics, occurring in a nonisothermal pellet with finite
external heat and mass transfer resistances (Wu et al., 1990a).
It is worth noting that optimization of the catalyst activity distribution is car-
ried out assuming that the support has a certain pore structure and hence specific
4 Introduction
effective diffusivities for the various components. Thus for a given pore structure,
the catalyst distribution within the support is optimized. An alternative optimiza-
tion in catalyst design is that of pore structure, while maintaining a uniform catalyst
distribution. In this case, the mass transport characteristics of the pellet are op-
timized. This approach has been followed by various investigators and has been
shown to lead to improvements in catalyst performance (cf. Hegedus, 1980; Pereira
et al., 1988; Hegedus and Pereira, 1990; Beeckman and Hegedus, 1991; Keil and
Rieckmann, 1994).
Much effort has also been invested in the preparation of nonuniformly active
catalysts. As insight is gained into the phenomena related to catalyst prepara-
tion, scientists are able to prepare specific nonuniform profiles. In this regard, it
should be recognized that catalyst loading and catalyst activity distributions are in
principle different characteristics. In catalyst preparation, the variable that is usu-
ally controlled is the local catalyst loading. However, under reaction conditions,
the local reaction rate constant is proportional to catalyst activity. The relation
between catalyst activity and catalyst loading is not always straightforward. For
structure-sensitive reactions, it depends on the particular reaction system, and
hence generalizations cannot be made. On the other hand, for structure-insensitive
reactions, catalyst activity is proportional to catalyst surface area. Thus, if the lat-
ter depends linearly on catalyst loading, then the catalyst activity and loading
distributions are equivalent. If the above dependence is not linear, then the two
distributions can be quite different. The majority of studies on nonuniform cat-
alyst distributions address catalyst activity optimization, although a few investi-
gators have considered catalyst loading optimization by postulating some type of
surface area–catalyst loading dependence (Cervello et al., 1977; Juang et al., 1981).
Along these lines, it was shown that when the relation between catalyst activity
and loading is linear, and the latter is constrained by an upper bound, the optimal
Dirac-delta distribution becomes a step distribution. However, if this dependence
is not linear, which physically means that larger catalyst crystallites are produced
with increased loading, then the optimal catalyst distribution is no longer a step,
but rather a more disperse distribution (Baratti et al., 1993). An important point
is that in order to make meaningful comparisons among various distributions, the
total amount of catalyst must be kept constant.
Work in the areas of design, performance, and preparation of nonuniform cat-
alysts has been reviewed by various investigators (Lee and Aris 1985; Komiyama
1985; Dougherty and Verykios 1987; Vayenas and Verykios, 1989; Komiyama and
Muraki, 1990; Gavriilidis et al., 1993a). In this monograph, these issues are dis-
cussed with emphasis placed on optimally distributed nonuniform catalysts. Spe-
cial attention is given to applications involving reactions of industrial importance.
1.3 Overview of Book Contents

This book is organized as follows. In Chapter 2, optimization of a single pellet is
addressed under isothermal and nonisothermal conditions. Both single and mul-
tiple reaction systems are discussed. Starting with simpler cases, the treatment is
1.3 Overview of Book Contents 5
extended to the most general case of an arbitrary number of reactions with arbi-
trary kinetics under nonisothermal conditions, in the presence of external trans-
port limitations. The analysis includes the effect of catalyst dispersion varying
with catalyst loading. Finally, the improved performance of nonuniform catalysts
is demonstrated through experimental studies for oxidation, hydrogenation, and
Fischer–Tropsch synthesis reactions.
Optimization of catalyst distribution in pellets constituting a fixed-bed reactor
requires one to take into account changes in fluid-phase composition and tem-
perature along the reactor. This is discussed in Chapter 3, for single and multiple
reactions, under isothermal and nonisothermal conditions. The discussion of ex-
perimental work is focused on catalytic oxidations.
Catalyst distribution influences the performance of systems undergoing deacti-
vation, and this issue is addressed in Chapter 4 for selective as well as nonselective
poisoning. Experimental work on methanation, hydrogenation, and NO reduction
is presented to demonstrate the advantages of nonuniform catalyst distributions.
In Chapter 5, the effect of catalyst distribution on the performance of inorganic
membrane reactors is discussed. In such systems, the catalyst can be located either
in pellets packed inside a membrane (IMRCF) or in the membrane itself (CMR).
Experimental results for an IMRCF are presented, and the preparation of CMRs
with controlled catalyst distribution by sequential slip casting is introduced.
In Chapter 6, special topics of particular industrial importance are discussed.
These include automotive catalysts, where various concepts of nonuniform dis-
tributions have been utilized; hydrotreating catalysts, which is a particular type
of deactivating system; composite catalysts, with more than one catalytic function
finding applications in refinery processes; biocatalysts; and functionalized polymer
resins, which find applications in acid catalysis.
The final Chapter 7 considers issues related to catalyst preparation. The discus-
sion is focused on impregnation methods, since they represent the most mature
technique for preparation of nonuniform catalysts. During pellet impregnation,
adsorption and diffusion of the various components within the support are im-
portant, and can be manipulated to give rise to desired nonuniform distributions.
The chapter concludes with studies where experimental results are compared with
model calculations.
2
Optimization of the Catalyst
Distribution in a Single Pellet
A
mong various reaction systems, investigation of optimal catalyst distribution
in a single pellet has received the most attention. Although the general prob-
lem of an arbitrary number of reactions following arbitrary kinetics occur-
ring in a nonisothermal pellet has been solved and will be discussed later in this
chapter, it is instructive to first consider simpler cases and proceed gradually to
the more complex ones. This allows one to understand the underlying physic-
ochemical principles, without complex mathematical details. Thus, we first treat
single reactions, under isothermal and nonisothermal conditions, and then analyze
multiple reactions.
2.1 The Case of a Single Reaction
2.1.1 Isothermal Conditions

In early studies, step distributions of catalyst were analyzed for the simple case of a
single reaction occurring under isothermal conditions. Researchers often treated
bimolecular Langmuir–Hinshelwood kinetics, which exhibits a maximum in the
reaction rate as a function of reactant concentration. Thus, there is a range of
reactant concentrations where reaction rate increases as reactant concentration
decreases. This feature occurs in many reactions; for example, carbon monoxide
or hydrocarbon oxidation, in excess oxygen, over noble metal catalysts (cf. Voltz
et al., 1973), acetylene and ethylene hydrogenation over palladium (Schbib et al.,
1996), methanation of carbon monoxide over nickel (Van Herwijnen et al., 1973),
and water-gas shift over iron-oxide-based catalyst (Podolski and Kim, 1974).
Wei and Becker (1975) and Becker and Wei (1977a) numerically analyzed the
effects of four different catalyst distributions. In three of these, the catalyst was
deposited in only one-third of the pellet: inner, middle, or outer (alternatively
called egg-yolk, egg-white, and eggshell, respectively). In the fourth it was uni-
formly distributed. The results are shown in Figure 2.1, where the effectiveness
factor η is shown as a function of the Thiele modulus φ. It may be seen that among
these specific distributions, for small values of φ (i.e. kinetic control) the inner
6
Figure 2.1. Isothermal effectiveness factor η as a function of

Thiele modulus φ for bimolecular Langmuir–Hinshelwood
kinetics in nonuniformly distributed flat-plate catalysts; di-
mensionless adsorption constant σ = 20. (From Becker and
Wei, 1977a.)
is best, while for large values of φ (i.e. diffusion control) the outer is best. For
intermediate values of the Thiele modulus, the middle distribution has the highest
effectiveness factor. So the question naturally arises: given a Thiele modulus φ,
among all possible catalyst distributions, which one is the best? This question can
be answered precisely, and is addressed next.
Definition of optimization problem

The optimization problem can be stated as follows: given a fixed amount of cat-
alytic material, identify the distribution profile for it within the support which
maximizes a given performance index of the catalyst pellet. In order to formulate
the problem in mathematical terms, the following equations are required: For a
single reaction
A → products (2.1)
the steady-state mass balance for a single pellet is given by

1 d dC
De n xn = a(x) r (C) (2.2)
x dx dx
where De is the effective diffusivity, x is the space coordinate, C is the reactant

concentration, r (C) is the reaction rate, and n is an integer characteristic of the
pellet geometry, indicating slab, cylinder, or sphere geometry for n = 0, 1, 2 re-
spectively. The catalyst activity distribution function a(x) is defined as the ratio
between the local rate constant and its volume-average value:
a(x) = k(x)/k̄ (2.3)

8 Optimization of the Catalyst Distribution in a Single Pellet
so that by definition

1
a(x) dVp = 1. (2.4)
Vp Vp
The boundary conditions (BCs) are

dC
x = 0: =0 (2.5a)
dx
x = R: C = Cf . (2.5b)
The constraint of a fixed total amount of catalyst means that k̄Vp is constant. In
dimensionless form, the above equations become

1 d n du
s = φ 2 a(s) f (u) (2.6)
s n ds ds
du
s = 0: =0 (2.7a)
ds
s = 1: u=1 (2.7b)
1
1
a(s)s n ds = (2.8)
0 n+1
where the following dimensionless quantities have been introduced:
u = C/Cf , s = x/R, φ 2 = r (Cf )R2 /De Cf .

(2.9)
f (u) = r (C)/r (Cf )
Since we are dealing with a single reaction, the catalyst performance is directly
related to the effectiveness factor, defined by
1
f (u)a(s)s n ds
η = 01 (2.10)
a(s)s n ds
0
which, using equation (2.8), yields

1
n + 1 du
η = (n + 1) f (u)a(s)s ds =
n
. (2.11)
0 φ2 ds s=1
Thus, the optimization problem consists in evaluating the catalyst distribution
a(s) which maximizes the effectiveness factor η under the constraints given by
equations (2.6)–(2.8).
Shape of optimal catalyst distribution

In order to proceed further, we need to know the specific form for the reaction
rate r (C). A variety of expressions can be used for this purpose. However, for
illustration we choose the bimolecular Langmuir–Hinshelwood kinetics,
r (C) = k̄C/(1 + KC)2 (2.12)

Figure 2.2. Shape of the dimensionless

bimolecular Langmuir–Hinshelwood rate
function f (u) = (1 + σ )2 u/(1 + σ u)2 , for
various values of the dimensionless adsorp-
tion constant σ . (From Morbidelli et al.,
1982.)
so that
r (C) (1 + σ )2 u
f (u) = = (2.13)
r (Cf ) (1 + σ u)2
where
σ = KCf . (2.14)
The shape of the rate function f (u) depends on the parameter σ and is shown in
Figure 2.2. In particular, f (u) has a unique maximum at
um = 1/σ. (2.15)
The dimensionless reaction rate reaches its maximum value in the range 0 < u < 1
for σ > 1, and at u = 1 for σ ≤ 1. Thus, summarizing, the Langmuir–Hinshelwood
kinetics exhibits a maximum value M at u = um , where
um = 1/σ, M = (1 + σ )2 /4σ for σ >1 (2.16a)
um = 1, M=1 for σ ≤1 (2.16b)
Since f (u) ≤ M, from the expression for η given by equation (2.11) it is evident that
1 1
η = (n + 1) f (u)a(s)s n ds ≤ (n + 1)M a(s)s n ds (2.17)
0 0
which, using equation (2.8), gives

η ≤ M. (2.18)
Therefore, for any activity distribution a(s), the effectiveness factor can never be
greater than M. It is apparent that if a function a(s) exists for which η = M, this
will constitute the solution of the optimization problem.
This function exists and is given by

δ(s − s̄)
a(s) = (2.19)
(n + 1) s̄ n
where δ(s − s̄) is the Dirac-delta function defined by
δ(s − s̄) = 0 for all s = s̄ (2.20a)
and
1
δ(s − s̄) ds = 1 (2.20b)
0
which physically corresponds to a sharp peak located at s = s̄. In our optimiza-

tion problem, s̄ is s̄ opt , the value of s where the rate function f (u) reaches its
maximum value; i.e., u( s̄ opt ) = um , where um is given by equation (2.16). In prac-
tice, this means that all the catalyst should be located at s = s̄ opt . By using the
Dirac-delta function property
1
f (s)δ(s − s̄) ds = f ( s̄) (2.21)
0
it can be easily shown that the activity distribution (2.19) is indeed the optimal
one. For this, equation (2.19) is substituted into equation (2.11) to give
1
δ(s − s̄ opt ) n
ηopt = f (u) n s ds = f (um ) = M. (2.22)
0 s̄ opt
Evaluation of optimal catalyst location

The evaluation of optimal catalyst location s̄ opt must be performed separately for
σ ≤ 1 and σ > 1.
If σ ≤ 1, then from (2.16b) um = 1, which is attained at the particle external
surface, and hence s̄ opt = 1. In this case, from equations (2.16) and (2.22), the
effectiveness factor is ηopt = 1.
If σ > 1, then some more computations are needed to evaluate the optimal
catalyst location. The details are available elsewhere (Morbidelli et al., 1982) and
lead to
4(σ − 1)
s̄ opt = 1 − for n = 0 (2.23a)
φ02

8(1 − σ )
s̄ opt = exp for n = 1 (2.23b)
φ02
φ02
s̄ opt = for n=2 (2.23c)
φ02 + 12(σ − 1)
where φ0 is a “clean” Thiele modulus which does not include the adsorption pa-
rameter σ and is defined as
φ02 = k̄R 2 /De = (1 + σ )2 φ 2 . (2.24)
Table 2.1. Location of optimal catalyst distribution and

corresponding effectiveness factor for isothermal,
bimolecular Langmuir–Hinshelwood kinetics, without
the existence of external transport resistances.
σ≤1 σ>1
s̄ opt 1 1 − a (n = 0)
exp(−) (n = 1)
1
(n = 2)
1+
(1 + σ )2
ηopt 1
4σ
a For > 1, one has s̄ opt = 0 and ηopt = f (ū), where ū is the
solution of 1 − ū − φ 2 f (ū) = 0. Here f (ū) is given by equation
(2.13).
The corresponding optimal effectiveness factor is
(1 + σ )2
ηopt = . (2.25)
4σ
The effect of all the involved physicochemical parameters on s̄ opt can be expressed
through a single dimensionless parameter , which is used in Table 2.1 to summa-
rize the results obtained so far:
4(n + 1)(σ − 1)
= . (2.26)
φ02
From inspection of equation 2.23 it appears that while for the infinite cylinder
(n = 1) and the sphere (n = 2) we have 0 ≤ s̄ opt ≤ 1 for all values of φ0 ≥ 0 and
σ ≥ 1, for the infinite slab (n = 0) the value of s̄ opt can become negative. This is
physically unrealistic, and in this case the optimal catalyst distribution is s̄ opt = 0,
i.e., the catalyst must be concentrated at the pellet center (Morbidelli et al., 1982).
The resulting value of the effectiveness factor will in this case be smaller than
ηopt given by equation (2.25), but it is still the maximum obtainable for the given
values of φ0 and σ .
From equations (2.23), it is seen that the catalyst location depends on the physic-
ochemical parameters of the system, i.e., the Thiele modulus φ0 and the adsorption
constant σ . On increasing the Thiele modulus or decreasing the adsorption con-
stant, the optimal location of the active catalyst moves from the interior of the
pellet to the external surface. This is shown in Figure 2.3, where for the spherical
pellet the s̄ opt -vs-φ0 curves for various σ values are plotted. Increasing the Thiele
modulus (keeping the adsorption constant unchanged) leads to larger diffusional
resistances, and therefore moves the location where u = um closer to the pellet’s
external surface. Similarly, decreasing the adsorption constant (keeping the Thiele
modulus unchanged) causes the maximum of the reaction rate to occur at larger
u values, which again are encountered closer to the external surface.
Figure 2.3. Optimal catalyst loca-

tion s̄opt as a function of Thiele mod-
ulus φ0 for various values of the di-
mensionless adsorption constant σ .
(From Morbidelli et al., 1982.)
Step catalyst distribution

From the practical point of view it is not possible, and from the sintering point
of view it is not desirable, to locate the catalyst as a Dirac-delta distribution. The
question therefore arises if the Dirac-delta distribution can be approximated by a
more convenient step-type distribution of narrow width. In this case, the catalyst
distribution is described by

0 for s < s1 or s > s2
a(s) = (2.27)
a for s1 < s < s2
where s1 = s̄ − , s2 = s̄ + , and a is a constant which is evaluated using con-

dition (2.8) as follows:
1
a= . (2.28)
s2n+1 − s1n+1
The system of equations (2.6)–(2.7) for the step distribution can only be solved
numerically. Care must be exercised in finding all possible solutions, since multiple
solutions may exist.
In Figure 2.4, the effectiveness factor is plotted as a function of Thiele modulus
for given values of σ, s̄, and . Note that the maximum value of the effectiveness
factor ηm is attained in a region of φ where multiplicity is present. This ηm value
is shown as a function of the step-distribution half thickness in Figure 2.5. As the
step width decreases, the behavior of the step distribution approaches that of the
Figure 2.4. Effectiveness factor η vs

Thiele modulus φ 2 for a slab pellet with
step distribution of catalyst centered at
s̄ = 0.8, and half thickness = 0.01. Bi-
molecular Langmuir–Hinshelwood re-
action, σ = 20. Here ηm is the maximum
value of the effectiveness factor, and
φm the corresponding Thiele modulus.
Figure 2.5. Maximum value of effectiveness

factor ηm for a step distribution of catalyst cen-
tered at s̄ = 0.5, as a function of step half thick-
ness . Bimolecular Langmuir–Hinshelwood
reaction, σ = 20. (From Morbidelli et al.,
1982.)
corresponding Dirac-delta distribution, i.e., as → 0, ηm → ηopt = 5.51 as given

by equation (2.25). For the parameter values considered in Figure 2.5, it can be seen
that if the thickness of the active layer is less than ∼5% of the pellet characteristic
dimension, the behavior of the two distributions is virtually the same. It is worth
noting that such widths can be realized experimentally, as discussed in section 7.2.
Sensitivity of catalyst performance to the step location

The sensitivity of catalyst performance to the step distribution location is of great
importance, due to the inevitable difficulties encountered experimentally in locat-
ing the catalyst at a precise point s̄ opt . This is illustrated in Figure 2.6, where the
effectiveness factor is plotted as a function of the location of the active zone center.
Figure 2.6. Values of the effectiveness factor η for a

pellet with step distribution of catalyst centered at
location s̄. Each curve is characterized by a value
of the Thiele modulus φ and by the correspond-
ing optimal location s̄opt . Bimolecular Langmuir–
Hinshelwood reaction, n = 0, σ = 20, = 0.025.
Table 2.2. Location of optimal catalyst distribution and corresponding

effectiveness factor for isothermal, bimolecular Langmuir–Hinshelwood kinetics
in the presence of external transport resistances.
σ>1
σ≤1 Bim > Bi∗ Bim < Bi∗

s̄ opt 1 1 − e a n=0 1
exp(−e ) n=1
1
n=2
1 + e
(1 + σ )2
ηopt f (ū) f (ū)
4σ
with ū given by with ū given by
φ f (ū)
2
φ 2 f (ū)
1 − ū − =0 1 − ū − =0
Bim (n + 1) Bim (n + 1)
a For Bim > Bi∗ /(1 − Bi∗ ) and Bi∗ < 1 we have s¯opt = 0, ηopt = f (ū), and ū is given by 1 − ū −
[(1 + Bim )/Bim ] φ 2 f (ū) = 0. Here f (ū) is given by equation (2.13).
If this is at s̄ < s̄ opt , then the effectiveness factor is lower than its maximum value
to an extent proportional to the difference s̄ − s̄ opt . However, if the location of
the active layer is s̄ > s̄ opt , then the effectiveness factor undergoes a dramatic de-
crease. The reason is related to the pellet’s multiplicity behavior. As illustrated in
Figure 2.4, a certain level of diffusional resistance is required to give the maximum
value of the effectiveness factor. If the actual diffusional resistance is larger, i.e.
φ > φm , the effectiveness factor on the upper branch undergoes a slight decrease,
while if it is smaller, i.e. φ < φm , it undergoes a dramatic decrease, since the upper
branch is no longer available. Thus, in preparing catalysts, if an error is anticipated,
it is better to err towards placing the step deeper within the pellet than s̄ opt .
Effect of external mass transfer resistance

In the presence of external mass transfer resistance, the pellet surface BC (2.7b)
changes to
du
s = 1: = Bim (1 − u) (2.29)
ds
where
Bim = kg R/De (2.30)
is the mass Biot number. In this case, the optimal activity distribution is again given
by a Dirac-delta function (Morbidelli and Varma, 1982). The effect of the exter-
nal mass transfer resistance on the optimal location is similar to that of internal
resistance. In particular, for decreasing values of Bim , the optimal location moves
towards the external surface of the pellet. It can be obtained from the equations
reported in Table 2.2, where the parameter e is given by
1 1
e = − + ∗ (2.31)
Bim Bi
where
φ02
Bi∗ = . (2.32)
4(n + 1)(σ − 1)
2.1.2 Nonisothermal Conditions

When the heat of reaction is high, the assumption of isothermality is not valid and
significant temperature gradients may develop. For typical gas–solid catalytic re-
actions, temperature gradients within the pellet are negligible in comparison with
those between the bulk fluid and external pellet surface (cf. Carberry, 1976). How-
ever, for pellets with nonuniform activity distribution, internal temperature gradi-
ents may become important. This has been demonstrated by Butt and coworkers
for nonuniform pellets resulting from deactivation (Kehoe and Butt, 1972; Butt
et al., 1977; Lee et al., 1978; Downing et al., 1979).

The case of arbitrary reaction kinetics will be taken up later, in section 2.1.3.
However, to illustrate the effect of temperature gradients, let us continue to con-
sider a single bimolecular Langmuir–Hinshelwood reaction (cf. Morbidelli et al.,
1985),
A → products. (2.33)
The steady-state mass and energy balances within the pellet under nonisothermal
conditions are

1 d dC
De n xn = a(x)r (C, T) (2.34)
x dx dx

1 d dT
λe n xn = −(−H) a(x) r (C, T) (2.35)
x dx dx
where λe is the effective thermal conductivity, T is the temperature, and −H is

the heat of reaction, while all the other symbols have been defined earlier.
The BCs are
dC dT
x = 0: = 0, =0 (2.36a)
dx dx
x = R: C = Cf, T = Tf . (2.36b)
The catalyst distribution a(x) for nonisothermal systems is defined as the ratio
between the local rate constant and its volume-average value, both evaluated at
the bulk fluid temperature:
a(x) = k(x, Tf )/k̄(Tf ). (2.37)

It is thus independent of temperature. In dimensionless form, the above equations

become

1 d n du
s = φ 2 a(s) f (u, θ) (2.38)
s n ds ds

1 d n dθ
s = −βφ 2 a(s) f (u, θ ) (2.39)
s n ds ds
du dθ
s = 0: = 0, =0 (2.40a)
ds ds
s = 1: u = 1, θ =1 (2.40b)
where
C x r (Cf , Tf )R2 r (C, T)
u= , s= , φ2 = , f (u, θ ) =
Cf R De Cf r (Cf , Tf )
(2.41)
T (−H)De Cf
θ= , β= .
Tf λe Tf
Combining equations (2.38) and (2.39), the following invariance can be obtained
(cf. Aris, 1975):

d n d
s (θ + βu) = 0 (2.42)
ds ds
Using the above relationship and the BCs (2.40), the dimensionless temperature
θ can be expressed as a function of dimensionless concentration u as follows:
θ = 1 + β(1 − u). (2.43)
This allows us to reduce the problem to a single differential equation

1 d n du
s = φ 2 a(s) f [u, 1 + β(1 − u)]. (2.44)
s n ds ds
Shape of optimal catalyst distribution

Equation (2.44) is the same as equation (2.6); therefore, following the same proce-
dure as in the isothermal case, it can be proven that the optimal distribution which
maximizes the effectiveness factor is again a Dirac-delta function (Morbidelli
et al., 1985). The only difference is that the dimensionless reaction rate now is

γβ(1 − u)
(1 + σ )2 u exp
1 + β(1 − u)
f (u) = 2 (2.45)
εβ(1 − u)
1 + σ u exp −
1 + β(1 − u)
where three new parameters appear: the dimensionless heat of reaction β, and
the activation energies γ and ε for the reaction rate constant and the adsorption
constant, given by
E −Ha
γ = , ε= . (2.46)
Rg Tf Rg Tf
Table 2.3. Location of optimal catalyst distribution and

corresponding effectiveness factor for nonisothermal,
bimolecular Langmuir–Hinshelwood kinetics, without
the existence of external transport resistances.
σ ≤ σc σ > σc
s̄ opt 1 1 − a n=0
exp(− ) n = 1
1
n=2
1 +
ηopt 1 f (um ),
where um is given by f (um ) = 0
a For > 1, one has s̄ opt = 0 and ηopt = f (ū), where ū is the solu-
tion of 1 − ū − φ 2 f (ū) = 0. Here f (ū) is given by equation (2.45),
and σc is given by equation (2.47).

The optimal performance of the catalyst pellet is obtained by locating the Dirac-
delta catalyst distribution so as to maximize the reaction rate by taking advantage
of both temperature and concentration gradients inside the pellet. Higher tem-
perature increases the reaction rate through a higher reaction rate constant, while
higher reactant concentration can either increase or decrease the rate of reaction,
depending upon the extent of reactant inhibition (see Figure 2.2).
The expressions for the optimal location and the corresponding effectiveness
factor are summarized in Table 2.3, where the critical value σc , which determines
if the optimal catalyst distribution is a surface or a subsurface one, is given by
1 − βγ
σc = . (2.47)
1 + β(γ + 2ε)
The effects of all physicochemical variables are described through a single dimen-
sionless parameter
(n + 1)(1 − um )
= (2.48)
φ 2 f (um )
where um is the value of u which maximizes the reaction rate f (u) given by equa-
tion (2.45).
In general, a larger dimensionless heat of reaction β leads to higher tempera-
tures within the pellet, thus increasing the ratio between reaction and diffusion
rates. This makes the reactant concentration profile steeper and therefore makes
the optimal location approach the external surface of the pellet (see Figure 2.7).
However, for small values of σ the optimal location can exhibit a minimum as
a function of β, as illustrated in Figure 2.7. As β increases, the optimal location
moves from the surface (at β = 0) towards the pellet interior, reaches a minimum
value, and then moves back to the pellet surface (β → ∞).
Figure 2.7. Optimal catalyst location s̄opt as a function of

the heat of reaction parameter β for various φ0 and σ val-
ues. Bimolecular Langmuir–Hinshelwood reaction, n = 2,
γ = 30, ε = 2.33. (From Morbidelli et al., 1985.)
Note that σ = 0 corresponds to a first-order reaction. In this case, under isother-

mal conditions, the reaction rate is maximized at the location where the reactant
concentration is the highest, i.e. at the external surface of the pellet. However, if
the pellet is nonisothermal, the optimal location can well be in the interior of the
pellet for a range of β values (see curve corresponding to σ = 0 in Figure 2.7).
More specifically, subsurface optimal catalyst locations are attained when σ > σc
(see Table 2.3), which for a first-order reaction leads to
βγ > 1. (2.49)
This behavior is solely due to the intrapellet temperature gradient, and indicates
that the temperature increase at the location of the reaction causes the net reaction
rate to increase sufficiently so as to more than overcome the adverse effect of its
decrease due to decreased reactant concentration. Due to the above temperature
gradient, effectiveness factors larger than unity can be attained.
2.1.3 Arbitrary Kinetics with External Transport Resistances

In the cases discussed above in sections 2.1.1 and 2.1.2, a single-variable (viz.
reactant concentration) optimization problem was solved. This strategy continues
to work for the case of a single reaction following arbitrary kinetics, as long as
either there are no external heat and mass transport resistances (Bim = Bih = ∞),
or with Bim = Bih when the resistances are finite. However, when Bim = Bih , a two-
variable optimization problem has to be solved. This problem has been addressed
through different approaches by Chemburkar et al. (1987) and by Vayenas and
Pavlou (1987a). Both studies showed that a Dirac-delta distribution remains the
optimal one. In this case the differential equations are reduced to a single equation
but with two variables as illustrated below.

The steady-state mass and energy balances within the pellet are again given by
equations (2.34) and (2.35). The BC at the pellet center also remains the same as
equation (2.36a). However, owing to the finite external mass and heat transport
resistances, the pellet surface BC (2.36b) now takes the form
dC dT
x = R: De = kg (Cf − C), λe = h(Tf − T). (2.50)
dx dx
The corresponding dimensionless equations are (2.38), (2.39), (2.40a), and
du dθ
s = 1: = Bim (1 − u), = Bih (1 − θ ) (2.51)
ds ds
where
Bim = kg R/De , Bih = hR/λe . (2.52)
From equations (2.38)–(2.40a) and (2.51), a relation between the dimensionless

concentration u and temperature θ at any location s can be obtained through
standard analysis (Aris, 1975). This gives
θ = 1 + βµ + βus (1 − µ) − βu (2.53)
where µ = Bim /Bih and the subscript s refers to the value at the pellet external
surface, i.e. s = 1. Using equation (2.53), the system of two differential equations
with variables u and θ reduces to one differential equation

1 d n du
s = φ 2 a(s)d(u, us ) (2.54)
s n ds ds
where d(u, us ) is given by f (u, θ) with θ replaced by (2.53). Note that two variables
u and us are present, since the value of us is not known a priori. For this reason, the
procedure used in previous cases (sections 2.1.1 and 2.1.2) cannot be applied here.
Details of the original analysis by Chemburkar et al. (1987) are not presented here,
because the simpler and more general proof in section 2.3.2 can also be applied
to this case.
Let us now simply report the final results. The optimal catalyst location is given
by
s̄ opt = 1 − e for n = 0 (2.55a)

s̄ opt = exp(−e ) for n=1 (2.55b)
s̄ opt = 1/(1 + e ) for n = 2 (2.55c)
where
1 uδ − um
e = · s . (2.56)
Bim 1 − uδs
Note that e is always positive, so that equations (2.55) always provide physically
meaningful values for s̄ opt , i.e. 0 < s̄ opt < 1. The only exception is for the slab
geometry (n = 0), where e can be larger than 1, in which case s̄ opt = 0. In order
to use the relation above we need to compute um and uδs . The first, um , is the value
of u which maximizes the function d(u, uδs ), while uδs is the smallest root of the
equation
(n + 1)Bim
1 − uδs = d um , uδs . (2.57)
φ2
This also provides the value of the optimal effectiveness factor, given by

ηopt = d um , uδs . (2.58)
The number of parameters upon which the active layer location depends is
now larger, and for the case of a nonisothermal mth-order reaction these are
the dimensionless heat of reaction β, the dimensionless activation energy γ , the
reaction order m, the geometry integer n, the mass Biot number Bim , the ratio
µ of mass to heat Biot numbers, and the Thiele modulus φ. These dimensionless
parameters in turn contain various physicochemical variables. However, given
a reaction, a fixed amount of catalyst, and a support, only two of them can be
considered as operating variables, viz. Tf and Cf . Any change in them affects
φ, β and γ , although the effect is not substantial for the latter two. Hence, the
dependence of the optimal location s̄ opt on the Thiele modulus φ is shown in
Figure 2.8 for typical values of the other parameters while one of them is varied
around its base value. The optimal catalyst activity location s̄ opt can exhibit a
discontinuity as a specific physicochemical parameter is varied, and this arises
from the fact that often multiple steady states are possible.
The effect of external transport resistances is illustrated in Figure 2.8.a and
b. As expected, when the resistances increase, i.e. when the mass or heat Biot
number decreases, the optimal location moves outwards. Also, the discontinuity
in the s̄ opt -vs-φ plot occurs at lower values of φ as the transport resistances in-
crease. The effect of pellet geometry (Figure 2.8.c) is that as n increases, the jump
occurs at a higher value of φ. It can also be seen in Figure 2.8.d–f that in general
as β increases (or γ increases, or m decreases): (1) the optimal location moves
inward for values of φ sufficiently small that no jump has occurred, and (2) the
jump occurs at smaller φ. However, for sufficiently small values of β or γ (or
sufficiently large values of m), the optimal location s̄ opt is always 1, regardless of
the value of the Thiele modulus. This behavior is directly related to the fact that
when
βγ
<1 (2.59)
m
for finite Bim and Bih , the optimal location for all φ is s̄ opt = 1 [compare with
equation (2.49)].
The discontinuity that the optimal catalyst location exhibits as a physicochemi-
cal parameter is varied results in a similar discontinuity in the effectiveness factor
Figure 2.8. Optimal location of catalyst s̄opt as a function of Thiele modulus φ for an mth-
order reaction. Bim = 50, µ = 10, n = 2, β = 0.05, γ = 25, and m = 1, unless otherwise
specified. (From Chemburkar et al., 1987.)
Figure 2.9. Effect of φ or σ on the max-

imum value of the effectiveness factor
ηm . Here Bim = 50, µ = 10, n = 2. (From
Chemburkar et al., 1987.)
(see Figure 2.9). When the jump occurs from a lower to a higher s̄ opt value, the
higher value is typically 1, i.e. at the external surface of the pellet (see Figure
2.8.e), and the corresponding effectiveness factor also jumps by some orders of
magnitude (see Figure 2.9.a). So, if there is free choice of the operating variables
(particularly the bulk fluid values Tf and Cf ), then the optimal catalyst location
can always be moved to the external surface of the pellet. In addition to the
convenience of preparing eggshell catalysts, such a strategy will also result in an
increased catalytic effectiveness.
Sensitivity of catalyst performance to catalyst location

Figure 2.9.a,b shows that there is a relatively narrow range of φ that gives a tremen-
dous enhancement in the effectiveness factor, and in the case of any deviations
from the set φ value, it is safer to have deviations leading to higher φ. Note that
Thiele modulus can be changed conveniently, for example by changing the bulk
fluid temperature Tf . The sensitivity of the effectiveness factor to φ is demon-
strated in Figure 2.10. For the given set of parameter values and φ = 0.56, the
optimum catalyst location is s̄ opt = 0.5. However, for this s̄ even a slight decrease
Figure 2.10. Effectiveness factor η of a pel-

let with a given Dirac-delta activity distri-
bution as a function of the Thiele modulus
φ. Bimolecular Langmuir–Hinshelwood
reaction, Bim = 50, µ = 10, n = 2, β = 0.03,
γ = 20, ε = 0, σ = 20. (From Chemburkar
et al., 1987.)
in the operating φ can cause a catastrophic decrease in the effectiveness factor.

Thus, from the practical point of view, it would be desirable to operate at a φ
somewhat higher than 0.56, giving a somewhat lower effectiveness factor on the
upper branch, but for which any small deviations from the operating φ on either
side do not cause a large deviation in the obtained effectiveness factor. This result
is similar to that discussed previously in the context of an isothermal reaction
(cf. Figures 2.4 and 2.6).
For bimolecular Langmuir–Hinshelwood reactions, the operating variables Tf
and Cf have significant effect not only on φ but also on σ . As shown in Figure 2.9.c,
there is no decrease in effectiveness factor when σ is increased beyond the value
at which the jump occurs. This indicates that the effectiveness factor is insensitive
to changes in σ , as long as the operating σ value is chosen sufficiently high.
2.1.4 Dynamic Behavior

When diffusion–reaction models of catalyst pellets contain nonlinearities due to
nonlinear reaction kinetics or thermal effects, their solutions may exhibit un-
usual behavior, such as multiplicity of solutions for steady state models, and self-
sustained oscillations for dynamic models (Aris, 1975). The case of pellets with
Dirac-delta distributions is no exception to this. Brunovska (1987, 1988) and Barto
et al. (1991) analyzed the steady-state and dynamic behavior of pellets with Dirac
and step-type distributions where a single reaction
A1 + ν2 A2 → products (2.60)
following Langmuir–Hinshelwood kinetics

k̄K1 K2 C1 C2
r= (2.61)
(1 + K1 C1 + K2 C2 )2
occurs.
Figure 2.11. Effectiveness factor η as

a function of the parameter φ 2 ln s̄.
Reaction A1 + ν2 A2 → products, fol-
lowing Langmuir–Hinshelwood ki-
netics according to equation (2.61),
n = 1, σ1 = 1, σ2 = 30, ν2 C f,1 /Cf,2 = 10,
De,1 /De,2 = 0.12. (From Brunovska,
1987.)
For Dirac-delta catalysts of cylindrical shape, the dependence of the effective-

ness factor on the parameter φ 2 ln s̄ is shown in Figure 2.11. One stable steady
state exists to the left of point A and to the right of point C. Between points B
and C there is a region of unique unstable steady states surrounded by a limit
cycle. Note that in the region between the two vertical lines the model exhibits
three steady states, of which only that on the upper branch is stable. It should be
noted that contrary to the steady-state solution, the transient solution depends
independently on the Thiele modulus φ and the catalyst location s̄ . Therefore the
stability character of the steady states in Figure 2.11 also changes with both φ and
s̄. In particular, it was found that the oscillatory region BC shrinks as the catalyst
location s̄ approaches the surface of the pellet.
Specifically, as the catalyst location moves towards the pellet external surface,
the self-sustained oscillations first decrease in amplitude and then disappear with
the formation of a stable steady state. The bifurcation diagram for Dirac-delta cat-
alysts forms an envelope contained within the solid lines in Figure 2.12. The lower
branch corresponds to point B and the upper branch to point C of Figure 2.11.
Within the envelope periodic oscillations were observed, while stable steady states
were obtained in the region outside the envelope. As noted above, as the catalyst
moves towards the external surface, the two branches merge, thus indicating that
oscillations are not possible.
Barto et al. (1991) further investigated this model using a different numerical
technique and revealed more complex dynamic behavior. The effect of deacti-
vation on the dynamics of Dirac pellets was also investigated. It was found that
the decrease of catalyst activity during deactivation has similar consequences to
the decrease of Thiele modulus for nondeactivating catalysts (Brunovska, 1987).
Starting from a point on the left of A (Figure 2.11) where only one stable steady
state exists, due to deactivation the region of periodic oscillations is reached. As
deactivation continues, oscillations die out and a region of unique stable steady
states is attained again, to the right of point C.
For step-type catalysts, the bifurcation diagram is somewhat deformed as com-
pared to the Dirac catalyst (dashed lines in Figure 2.12). For a fixed catalyst loca-
tion, the values of the lower and upper limits of the Thiele modulus within which
periodic oscillations occur both increase from Dirac to step catalyst. It was also
found that while for a Dirac-delta distribution the value of the Thiele modulus
Figure 2.12. Bifurcation diagram. Solid curve, Dirac-delta

activity distribution; dashed curve, step-type distribu-
tion. Reaction A1 + ν2 A2 → products follows Langmuir–
Hinshelwood kinetics according to equation (2.61), n = 1,
= 0.1, σ1 = 1, σ2 = 30, ν2 Cf,1 /Cf,2 = 10, De,1 /De,2 = 0.12.
Periodic oscillations are observed within the envelopes.
(From Brunovska, 1988.)
does not affect the amplitude of oscillations significantly, for a step distribution
the amplitude increases as the Thiele modulus decreases. Furthermore, while for
the Dirac distribution the relation of steady-state effectiveness factor to φ 2 ln s̄
is independent of catalyst location s̄ (see Figure 2.11), for step-type catalysts it
depends on both step location and width.
2.2 Multiple Reactions

When more than one reaction is involved, other catalyst performance indexes
related to the production of desired species become important in addition to the
effectiveness factor, viz. selectivity and yield. The case where two reactions, either
consecutive or parallel, following arbitrary kinetics occur was treated by Vayenas
and Pavlou (1987b) to determine the catalyst distribution which maximizes selec-
tivity. In all cases, the optimal distribution is an appropriately located Dirac-delta
function.
Parallel reactions: A1 → A2 , A1 → A3
For parallel reactions that follow power-law kinetics the optimal location of the
Dirac-delta is at the surface of the pellet if the order of the desired reaction is
higher than the order of the undesired one; in the opposite case it is at the center
of the pellet. Egg-white distributions can be optimal for more complex kinetics.
As an example, in Figure 2.13 we consider the case where the desired reaction
follows bimolecular Langmuir–Hinshelwood kinetics, while the undesired reac-
tion follows unimolecular Langmuir–Hinshelwood kinetics, and the two involve
different adsorption sites. The dependence of the optimal catalyst location, se-
lectivity, and dimensionless reaction rate of the desired product on the ratio of
Thiele moduli, κ = (φ2 /φ1 )2 , is shown for the optimal Dirac, uniform, and exter-
nal shell catalyst distributions. The performance index used for optimization was
selectivity. When the undesired reaction becomes faster (i.e., κ increases), the se-
lectivities for all three types of distributions decrease. However, the Dirac-type
catalyst always gives the highest selectivity value. Simultaneously, the location of
the Dirac distribution moves towards the external surface of the pellet. The net
rate of production for the desired product is also consistently the highest for the
Dirac distribution. It is worth noting that the sensitivity of selectivity to catalyst
location behaves qualitatively in the same manner as discussed previously for the
effectiveness factor in the case of a single reaction.
Figure 2.13. Effect of the ratio of Thiele moduli, κ, on the optimal catalyst
location, s̄opt (solid lines), global selectivity S (dashed lines), and dimen-
sionless rate of production of the desired product A2 , f¯ (dotted lines),
for the optimal distribution (Sopt , f¯opt ), uniform distribution for slab ge-
ometry (Sun , f¯un ), and external shell distribution (Sext, f¯ext ). Parallel reac-
1 2
tions: A1 −→ A2 , A1 −→ A3 ; f1 (u1 ) = (1 + σ1 )2 u1 /(1 + σ1 u1 )2 , f2 (u1 ) =
(1 + σ2 )u1 /(1 + σ2 u1 ), σ1 = 5, σ2 = 20, φ1 [(n + 1)/gv ]1/2 = 1, De,1 /De,2 = 1.
Optimization is based on selectivity. (From Vayenas and Pavlou, 1987b.)
Consecutive reactions: A1 → A2 → A3
When two consecutive reactions occur, where the desired one follows bimolecular
Langmuir–Hinshelwood kinetics and the undesired one follows linear kinetics, the
Dirac catalyst distribution again gives the highest selectivity for all values of κ,
as illustrated in Figure 2.14. The difference in selectivity obtained with different
distributions (Sopt , Sun , and Sext ) increases with κ. The optimal distribution not only
gives the highest selectivity, which is the optimization index, but also the highest
net rate of production of the desired species. However, it should be noted that the
optimal catalyst location for selectivity maximization is in general different from
the optimal catalyst location for maximum production of the desired product.
For the reaction system considered, the catalyst location is rather insensitive to
changes in κ, except for sufficiently high values of κ, where the optimal location
moves rapidly towards the center of the pellet, because diffusional resistances
slow down the increasingly faster first-order undesired reaction. Although a Dirac-
delta function is always the optimal distribution, the actual performance of Dirac
catalysts will depend strongly on the type of kinetics that the reactions follow, and
can be quite different than the specific cases discussed above.
Figure 2.14. Effect of the ratio of Thiele moduli, κ, on the optimal catalyst
location s̄opt (solid lines), global selectivity, S (dashed lines), and dimen-
sionless rate of production of the desired product A2 , f¯ (dotted lines), for
the optimal distribution (Sopt , f¯opt ), uniform distribution for slab geometry
(Sun , f¯un ), and external shell distribution (Sext , f¯ext ). Consecutive reactions:
1 2
A1 −→ A2 −→ A3 ; f1 (u1 ) = (1 + σ )2 u1 /(1 + σ u1 )2 , f2 (u2 ) = u2 , σ = 8,
φ1 [(n + 1)/gv ]1/2 = 1, De,1 /De,2 = 1, uf,2 = 1. Optimization is based on
selectivity. (From Vayenas and Pavlou, 1987b.)

The improved performance of Dirac catalysts, as demonstrated in previous sec-
tions, arises from the appropriate manipulation of concentration gradients which
develop within the pellet. When detrimental, they are avoided by positioning the
catalyst on the surface of the pellet. When they are beneficial, however, one can
take advantage of them by placing the catalyst at an appropriate location inside
the pellet. Temperature gradients are also manipulated in a similar way. However,
their effect is generally more pronounced because of the exponential dependence
of reaction rates on temperature. Similar to the isothermal case, parallel, consec-
utive, and also triangular reaction networks, occurring in nonisothermal pellets,
show optimum performance when the catalyst is located at a specific position
which depends upon the physicochemical parameters of the system (Morbidelli
et al., 1984a; Vayenas and Pavlou, 1988; Vayenas et al., 1989; Pavlou and Vayenas,
1990a; Letkova et al., 1994). In this case, the activation energies and heats of
reaction are also involved.
Parallel reactions: A1 → A2 , A1 → A3
As an example consider two parallel reactions, where the desired one is exother-
mic, has low activation energy, and exhibits second-order kinetics, while the unde-
sired one is endothermic, has high activation energy, and exhibits linear kinetics
(Vayenas and Pavlou, 1988). The effect of the ratio of Thiele moduli, κ = (φ2 /φ1 )2 ,
on the optimal location and selectivity, where the latter is optimized, is shown in
Figure 2.15. In order to obtain a physical understanding of this system, it should be
Figure 2.15. Effect of the ratio of Thiele moduli,

κ, on the optimal catalyst location s̄opt (solid lines),
optimal global selectivity Sopt , and global selectivity
obtained with uniform (Sun ) and external shell (Sext )
1
catalyst distributions. Parallel reactions: A1 −→ A2
2
(second order), A1 −→ A3 (first order); desired
product A2 ; β1 = 0.1, β2 = −0.7, γ1 = 5, γ2 = 25, φ1 [(n +
1)/gv ]1/2 = 1. Optimization is based on selectivity.
(From Vayenas and Pavlou, 1988.)
realized that increasing reactant concentration and decreasing temperature favor

the desired reaction. For small κ values, the rate of the undesired reaction is rela-
tively small and the desired exothermic reaction creates higher temperatures in-
side the pellet, if the catalyst is located there. This however has a detrimental effect
on selectivity because of the relative activation energies, which adds to the detri-
mental concentration effect. Therefore, the optimum catalyst location is at the sur-
face of the pellet. For large κ values, the undesired reaction is fast and overcomes
the thermal effect of the exothermic desired reaction. As a result, if the catalyst
is located inside the pellet, the temperature decreases, which benefits the desired
reaction. This thermal effect easily surmounts the always negative concentration
effect, and consequently the optimum catalyst location moves inside the pellet.
Consecutive reactions: A1 → A2 → A3
The case of two consecutive exothermic reactions with linear kinetics and higher
activation energy for the first desired reaction is illustrated in Figure 2.16 (Vayenas
et al., 1989). Since the reaction orders are the same for both reactions, in this case
it is clear that concentration gradients do not affect selectivity at all. Increased
temperature favors the desired reaction, since its activation energy is larger than
Figure 2.16. Effect of the ratio of Thiele moduli κ on the

optimal catalyst location s̄opt (solid lines), optimal global se-
lectivity Sopt , and global selectivity obtained with uniform,
(Sun ) and external shell (Sext ) catalyst distributions. Consec-
1 2
utive reactions: A1 −→ A2 (first order), A2 −→ A3 (first or-
der); desired product A2 ; β1 = 0.4, β2 = 0.1, γ1 = 10, γ2 = 5,
φ1 [(n + 1)/gv ]1/2 = 1, De,1 /De,2 = 1, uf,2 = 1. Optimization is
based on selectivity. (From Vayenas et al., 1989.)
that of the undesired one, i.e. γ1 > γ2 . Such temperatures are obtained by placing
the catalyst inside the pellet. For high values of κ, which indicates that the unde-
sired reaction is so fast that it approaches diffusion control, the optimal catalyst
location moves rapidly towards the center of the pellet.
A system of multiple reactions of great industrial importance, the ethylene
epoxidation reaction network, was investigated by Morbidelli et al. (1984a). Ethy-
lene reacts over silver catalyst to give ethylene oxide, which is the desired product,
and carbon dioxide, which is the undesired by-product. Ethylene oxide is also com-
busted, but this reaction is less important than the combustion of ethylene, and so
it was not included in the above study. It was shown that the selectivity to ethylene
oxide is maximized when the catalyst is located at the external surface of the pel-
let. This is expected for parallel exothermic reactions with similar kinetics when
the undesired reaction has a higher activation energy than the desired one. The
maximum selectivity is attained at the surface of the pellet, even if combustion
of ethylene oxide is included (Pavlou and Vayenas, 1990a). However, if the net
production rate of ethylene oxide is to be maximized, then subsurface locations
are optimal, a fact which arises from the nonisothermality of the pellet.
Numerical optimization
A numerical optimization technique has been developed by Baratti et al. (1990),
and applied to the consecutive–parallel reaction network
A1 → A2 → A3
(2.62)
A1 → A4 .
The technique is based on the method of orthogonal collocation over finite el-
ements, whereby the differential equations are reduced to a set of nonlinear
algebraic equations, whose solution is obtained through the Newton–Raphson
method. It was shown that the number of collocation points used in the calcula-
tions is critical, and if it is too small, then erroneous results can arise. In Figure
2.17, the numerically calculated optimal activity distribution is shown for the re-
acting system noted above, where all the reactions are exothermic and follow
first-order kinetics. The optimization index is the pellet effectiveness factor, and
it corresponds to maximum consumption rate of component A1 . The shape of the
obtained distribution closely resembles a Dirac delta function, in full agreement
with analytical calculations. This numerical technique will also be used in the opti-
mization of complex reacting systems, and will be discussed further in section 2.5.4.
2.3 The General Case of a Complex Reaction System

Optimization methods based on the analysis of the performance-index integral
[e.g. equation (2.11)] and of the reaction rate function f (u) cannot be readily
extended to complex reaction systems, which are frequently encountered in in-
dustrial applications. For this purpose, a method has been developed based on a
necessary condition for optimality using variational techniques (Wu et al., 1990a;
Figure 2.17. Optimal catalyst activity distribution for effectiveness

factor maximization. Consecutive–parallel first-order reactions: φ1 =
0.5, φ2 = 1, φ3 = 0.5, β1 = 0.2, β2 = 0.12, β3 = 0.18, γ1 = 20, γ2 = 10,
γ3 = 20, De,1 /De,2 = 1, uf,2 = 0. (From Baratti et al., 1990.)
Morbidelli et al., 1991). Before proceeding to analyze complex reaction systems,

it is first illustrated for the simple case of a single isothermal reaction.
2.3.1 An Illustrative Example

As described in section 2.1.1, the relevant dimensionless equations for a single
isothermal reaction with no external mass transfer resistance are as follows. The
reactant mass balance gives

1 d du
L[u] = n sn = φ 2 a(s) f (u) (2.63)
s ds ds
along with BCs
du
s = 0: =0 (2.64a)
ds
s = 1: u = 1. (2.64b)
From the definition of the activity distribution function,
a(s) = k(s)/k̄, (2.65)
while the constraint on a(s) arising from its definition requires

1
(n + 1) a(s)s n ds = 1. (2.66)
0
The effectiveness factor is given by

1
n + 1 du
η = (n + 1) f (u)a(s)s n ds = . (2.67)
0 φ2 ds s=1
The optimization problem consists in finding the function a(s) which maximizes
the performance index η under the constraint given by equations (2.63), (2.64),
and (2.66).
The necessary condition for optimality

Let us assume that a ∗ (s) is the optimal catalyst distribution which maximizes the
objective functional (2.67). Any other distribution a(s) in the neighborhood of
a ∗ (s) can then be expressed as a perturbation about a ∗ (s):
a(s) = a ∗ (s) + δa (2.68)
where both a(s) and a ∗ (s) satisfy the constraint (2.66). The dimensionless con-
centration profile u(s), resulting from a(s), can also be represented as a per-
turbation about the optimal concentration profile u∗ (s) resulting from a ∗ (s) as
follows:
u(s) = u∗ (s) + δu. (2.69)
If the perturbation is chosen sufficiently small, then a first-order expansion about

a ∗ is sufficient, and linearization of equation (2.63) about a ∗ yields
L[δu] = φ 2 (a ∗ f ∗ δu + f ∗ δa) (2.70)
with the BCs

dδu
s = 0: =0 (2.71a)
ds
s = 1: δu = 0 (2.71b)
where

d f (u)
f ∗ = f (u∗ ) and f ∗ = . (2.72)

du
u=u∗
Similarly, for the effectiveness factor,
1
δη = η[a(s)] − η[a ∗ (s)] = (n + 1) (a ∗ f ∗ δu. + f ∗ δa)s n ds. (2.73)
0
In order to obtain the complete influence of the perturbation δa on the objective

functional, we need to adjoin the linearized constraint (2.70) to the objective
functional by introducing a Lagrange multiplier λ(s). This is done by adding the
identity
1
(n + 1) λ{L[δu] − φ 2 (a ∗ f ∗ δu + f ∗ δa)}s n ds = 0 (2.74)
0
to equation (2.73), to get

1
δη = (n + 1) {(1 − φ 2 λ) f ∗ δa + (1 − φ 2 λ)a ∗ f ∗ δu + λL[δu]}s n ds. (2.75)
0
Since the Lagrange multiplier is arbitrary, we can define it as follows:
L[λ] = (φ 2 λ − 1)a ∗ f ∗ (2.76)
with BCs
dλ
s = 0: =0 (2.77a)
ds
s = 1: λ = 0. (2.77b)
The last term on the right-hand side of equation (2.75), when integrated twice by
parts, becomes
1 1
λL[δu]s ds =
n
δu L[λ]s n ds (2.78)
0 0
Substituting the above relation into equation (2.75) and using equations (2.76)
and (2.77) leads to
1
δη = (n + 1) G ∗ δa s n ds (2.79)
0
where
G ∗ = (1 − φ 2 λ) f ∗
(2.80)
will be referred to subsequently as the G function. The asterisk indicates that

all quantities in the G function are evaluated at the assumed optimal activity
distribution, a ∗ (s).
Equation (2.79) indicates the direct influence of δa on δη. The necessary condi-
tion for optimality is that δη ≤ 0 for all possible small variations δa. Using equation
(2.68), this becomes
1 1
G ∗ a ∗ (s)s n ds ≥ G ∗ a(s)s n ds. (2.81)
0 0
Shape of the optimal catalyst distribution

Having assumed that a ∗ (s) is the optimal distribution, we prove that a ∗ (s) can sat-
isfy the necessary condition (2.81) for optimality only if it is substantially nonzero
solely at the point s̄ where the corresponding G ∗ (s) attains its maximum value
over 0 < s < 1. By substantially nonzero we mean that
s+ε
¯
a ∗ (s)s n ds = 0 (2.82)
s−ε
¯
for arbitrary small ε > 0. The above statement is proven by contradiction, i.e. by
showing that if it is not true, and a ∗ (s) is substantially nonzero at least at one point
s = s̄ where
G ∗ (s ) < G ∗ (s̄) (2.83)
then the catalyst distribution a ∗ (s) cannot be optimal.

For this purpose, let us define a new distribution a(s), which is a small pertur-
bation of a ∗ (s), satisfying the constraint (2.66):
 ∗ n ∗
a (s) + γ [s + s − s̄] a (s + s − s̄)/s for s̄ − ε < s < s̄ + ε
 n
a(s) = (1 − γ )a ∗ (s) for s − ε < s < s + ε


 ∗
a (s) otherwise.
(2.84)
Since γ > 0 is arbitrarily small, a(s) can be made infinitesimally close to a ∗ (s),
as required by equation (2.68). Therefore the necessary condition for optimality,
(2.81), can be used. From equation (2.84) it follows that
1 s̄+ε
G ∗ (s)a(s)s n ds = γ [s + s − s̄]n a ∗ (s + s − s̄)G ∗ (s) ds
0 s̄−ε

+ a ∗ (s)G ∗ (s)s n ds
(0,1)\(s −ε,s +ε)
s +ε
+ (1 − γ ) a ∗ (s)G ∗ (s)s n ds. (2.85)
s −ε
From equation (2.83) and the continuity of G ∗ (s), it follows that for sufficiently
small ε > 0
G ∗ (s − s + s̄) > G ∗ (s) for s − ε < s < s + ε. (2.86)
Using this equation, we determine a lower bound on the first integral on the right-
hand side of equation (2.85). For this, it is convenient to introduce the new variable
z = s + s − s̄, and then from equation (2.86) we obtain
s+ε
¯ s +ε
n ∗ ∗
[s + s − s̄] a (s + s − s̄)G (s) ds > G ∗ (s)a ∗ (s)s n ds. (2.87)
s̄−ε s −ε
Using equation (2.87), equation (2.85) yields

1 1
∗
G (s)a(s)s ds >
n
G ∗ (s)a ∗ (s)s n ds (2.88)
0 0
which violates the necessary condition (2.81) for optimality. Since the distribution
a(s), which is a small perturbation about a ∗ (s), violates (2.81), a ∗ (s) cannot be
optimal. Therefore, in order for a ∗ (s) to be optimal, it has to be substantially
nonzero only at the point s̄ where the corresponding G ∗ (s) is maximum, which
means that a ∗ (s) must be a Dirac-delta function. Although it is highly unlikely that
G ∗ (s) will attain its maximum simultaneously at more than one point, it should
be mentioned that in this case the optimal distribution would take the form of
multiple Dirac-delta functions (Pavlou et al., 1991; Morbidelli et al., 1991; see also
section 2.5.5).
It is worth noting that the above proof fails when the optimal activity distribu-
tion a ∗ (s) generates a function G ∗ (s) which is constant and equal to its maximum
value Gm∗ for all values of s where a ∗ (s) is nonzero:

G ∗ (s) = Gm∗ for a ∗ (s) > 0
(2.89)
G ∗ (s) ≤ Gm∗ for a ∗ (s) = 0.
In this case, it is not possible to devise a distribution a(s), infinitesimally close to
a ∗ (s), which violates the necessary condition (2.81). It is in fact impossible to move
an infinitesimally small portion of a ∗ (s) from an interval (s − ε, s + ε) to another
(s − ε, s + ε) where G ∗ (s) is larger, as we did in equation (2.84). Therefore, in
this particular case, the optimal distribution does not need to exhibit the shape
of a Dirac delta function. Although this situation is rather unlikely in practice, at
least one example exists (Keller et al., 1984), which will be discussed in section
5.3, since it refers to a catalytic membrane.
2.3.2 General Reaction System

The above procedure is now extended to a general reaction system comprising an
arbitrary number of reactions with arbitrary kinetics, occurring in a nonisothermal
pellet with finite external mass and heat transfer resistances.

Consider a general reaction scheme where J independent reactions occur and I
components (A1 , A2 , . . . , AI ) are involved:
I
νij Ai = 0 ( j = 1, . . . , J ).
i=1
Each reaction follows arbitrary kinetics, hence

r j = r j (C1 , C2 , . . . , CI , T) ≡ r j (c, T), (2.90)
where c is the vector of concentrations. The steady-state mass and energy balances
in a porous catalyst pellet, allowing for external transport resistances, are given by
J
1 d n dCi
De,i n x = a(x) νij r j (c, T) (i = 1, . . . , I ) (2.91)
x dx dx j=1
J
1 d dT
λe n xn = −a(x) (−Hj )r j (c, T) (2.92)
x dx dx j=1
along with the BCs

dCi
x = 0: =0 (i = 1, . . . , I) (2.93a)
dx
dCi
x = R: De,i = kg,i (Cf,i − Ci ) (i = 1, . . . , I) (2.93b)
dx
dT
x = 0: =0 (2.93c)
dx
dT
x = R: λe = h(Tf − T). (2.93d)
dx
In dimensionless form, the above equations become

J
1 d dui
L[ui ] = n sn = a(s)Di νij φ 2j f j (u, θ ) (i = 1, . . . , I) (2.94)
s ds ds j=1
J
1 d dθ
L[θ ] = n sn = −a(s) β j φ 2j f j (u, θ ) (2.95)
s ds ds j=1
dui
s = 0: =0 (i = 1, . . . , I) (2.96a)
ds
dui
s = 1: = Bim,i (uf,i − ui ) (i = 1, . . . , I) (2.96b)
ds
dθ
s = 0: =0 (2.96c)
ds
dθ
s = 1: = Bih (1 − θ ) (2.96d)
ds
where
ui = Ci /Cf,1 , s = x/R, Di = De,1 /De,i
φ 2j = r j (Cf,1 , Cf,1 , . . . , Cf,1 , Tf )R2 /De,1 Cf,1
f j (u, θ) = r j (C1 , C2 , . . . , CI , T)/r j (Cf,1 , Cf,1 , . . . , Cf,1 , Tf ) (2.97)
θ = T/Tf , β j = (−Hj )De,1 Cf,1 /λe Tf
Bim,i = kg,i R/De,i , Bih = hR/λe
and the activity distribution a(s) satisfies constraint (2.66). Several indexes can
be used to evaluate the performance of a catalyst pellet when multiple reactions
occur. These include the effectiveness factor, which represents the overall con-
sumption rate of a certain reactant; the selectivity, which represents the fraction
of a given reactant to a desired product; and the yield, which represents the overall
production rate of a certain product.
The effectiveness factor with respect to the Mth component is defined as the
ratio between the actual consumption rate of the component and the rate which
would have existed in the absence of all transport resistances:
1 J n
(n + 1) 0 a(s) j=1 ν Mj φ j f j (u, θ ) s ds
2
ηM = J . (2.98)
j=1 ν Mj φ j
2
The selectivity of the Mth component (reactant) towards the Nth component
(product) is defined as the ratio between the rate of production of component N
and the rate of consumption of component M:
1 J n
j=1 ν Nj φ j f j (u, θ ) s ds
2
0 a(s)
SNM = 1 J . (2.99)
j=1 ν Mj φ j f j (u, θ ) s ds
2 n
0 a(s)
The yield of desired product N with respect to reactant M is defined as the ratio
between the actual rate of production of component N and the rate of consump-
tion of reactant M which would have prevailed in the absence of all transport
resistances:
1 J n
(n + 1) 0 a(s) j=1 ν Nj φ j f j s ds
2
YNM = η M SNM = J (2.100)
j=1 ν Mj φ j
2
In order to simplify the notation, let us introduce the (I + 1)-dimensional vec-

tors
 
J
 D1 ν1 j φ 2j f j (x) 
 
 j=1 
   J   
u1   1
 D2 ν2 j φ 2j f j (x) 
 u2     uf,2 
   j=1   
 ..   ..   .. 
x =  . , w(x) =  .
,
 x =   (2.101)
 . 
f
   
 uI   J   uf,I 
 DI ν I j φ 2j f j (x) 
θ   1
 
 j=1

 J 
 − β φ 2 f (x) 
j j j
j=1
and the diagonal matrix

 
Bim,1 0
 0 Bi 0 
 m,2 
 .. .. .. 
A= . . . . (2.102)
 
 0 Bim,I 0 
0 Bih
The basic equations (2.94)–(2.96) can now be written concisely as follows:

L[x] = a(s)w(x) (2.103)
with BCs
dx
s = 0: =0 (2.104a)
ds
dx
s = 1: = A(xf − x). (2.104b)
ds
The expressions for the performance indexes [equations (2.98)–(2.100)] take the
form
1
(n + 1) 0 a(s)w M (x)s n ds
ηM = (2.105)
DM Jj=1 ν Mj φ 2j
1
DM 0 a(s)w N (x)s n ds
SNM = (2.106)
DN 1 a(s)w M (x)s n ds
0
1
(n + 1) a(s)w N (x)s n ds
YNM = η M SNM = J
0
. (2.107)
DN j=1 ν Mj φ 2j
Since the expressions for the effectiveness factor (2.105) and yield (2.107) have
the same form, the objective function in these cases, up to a constant multiplier,
can be written in general as
1
U1 = (n + 1) a(s)wi (x)s n ds. (2.108)
0
Similarly, for the selectivity the objective function is given by

1
a(s)wi (x)s n ds
U2 = 01 . (2.109)
n
0 a(s)w j (x)s ds
The optimality conditions

The optimality conditions are obtained as in section 2.3.1 by linearizing the con-
straints (mass and energy balances) and the objective functional and adjoining
them with an (I + 1)-dimensional Lagrange multiplier vector. The G functions (G1
for effectiveness factor and yield, and G2 for selectivity) obtained are (Wu et al.,
1990a)
G1∗ = wi − λT w (2.110)
j wi − i w j − λ w T
G2∗ = (2.111)
2j
where
1
k = (n + 1) a ∗ (s)wk s n ds. (2.112)
0
The necessary conditions for optimality of effectiveness factor and yield are
1 1
∗ ∗
G1 a (s)s ds ≥
n
G1∗ a(s)s n ds (2.113)
0 0
and for selectivity

1 1
G2∗ a ∗ (s)s n ds ≥ G2∗ a(s)s n ds. (2.114)
0 0
We may observe that, even though the expressions for the G function are different
for the two optimization problems, the form of the optimality condition is the same.
The method utilized in the case of a single isothermal reaction to prove that the
optimal catalyst distribution is a Dirac-delta function is based on the existence
and form of the G function. Since the optimality conditions (2.113) and (2.114)
have the same form as (2.81), and furthermore, no matter how complicated the G
functions may be, when a ∗ is given, G1∗ and G2∗ are functions only of s, the same
arguments can be followed also in this case.
Thus, it can be concluded that, for any catalyst performance index (i.e. effec-
tiveness, selectivity, or yield) and for the most general case of an arbitrary number
of reactions, following arbitrary kinetics, occurring in a nonisothermal pellet, with
finite external mass and heat transfer resistances, the optimal catalyst activity distri-
bution is a Dirac-delta function.
An alternative method to prove that the optimal distribution is a Dirac-delta

function has been reported by Ye and Yuan (1992).

Having established that the optimal distribution is a Dirac-delta function, it is now
only required to determine its location. This can be conveniently done by seeking
directly the Dirac-delta location which maximizes the desired performance index,
as described below.
For a Dirac catalyst distribution
δ(s − s̄)
a(s) = (2.115)
(n + 1) s̄ n
the effectiveness factor with respect to the Mth component (2.98) reduces to the
algebraic equation
J
j=1 ν Mj φ j f j (ū, θ̄)
2
ηM = J , (2.116)
j=1 ν Mj φ j
2
where ū and θ̄ represent the dimensionless species concentrations and temperature

at the Dirac location s̄. Similarly, the selectivity (2.99) of reactant M towards
product N reduces to
J
j=1 ν Nj φ j f j (ū, θ̄ )
2
SNM = J , (2.117)
j=1 ν Mj φ j f j (ū, θ̄ )
2
and the yield (2.100) towards the desired product N with respect to reactant M to
J
j=1 ν Nj φ j f j (ū, θ̄ )
2
YNM = η M SNM = J . (2.118)
j=1 ν Mj φ j
2
In addition, the diffusion–reaction equations (2.94)–(2.96) reduce to a set of alge-

braic equations

1 Di Jj=1 νij φ 2j f j (ū, θ̄ )
ūi − uf,i + − ψn (s̄) =0 (2.119)
Bim,i n+1
J
j=1 β j φ j f j (ū, θ̄ )
2
1
θ̄ − 1 − − ψn (s̄) =0 (2.120)
Bih n+1
where

 s̄ − 1 for n = 0
ψn (s̄) = ln s̄ for n = 1 (2.121)

1 − 1/ s̄ for n = 2.
For a given catalyst location s̄, equations (2.119) and (2.120) are solved for ū and θ̄
and then the performance index of interest is calculated from equations (2.116)–
(2.118). Caution should be exercised during the numerical solution, because the
nonlinearity of the equations implies the possibility of multiple solutions. The op-
timal catalyst location s̄ opt is obtained by identifying the s̄ value in the interval
[0, 1] which maximizes the desired performance index. Note that for a given re-
acting system, the optimal catalyst location s̄ opt depends strongly on whether the
performance index considered is effectiveness, selectivity, or yield.
2.4 Catalyst Dispersion Considerations

In sections 2.1–2.3, we have been concerned with optimization of the catalyst ac-
tivity distribution. However, when preparing a catalyst we have control over the
catalyst loading distribution. Therefore it is desirable to use catalyst loading di-
rectly as the optimization variable. For this, a relationship between catalyst activity
and loading is needed. For structure-insensitive reactions, the activity of a catalyst
is proportional to its surface area. Hence, it suffices to know the dependence of its
surface area on its loading. At sufficiently low catalyst concentrations, the surface
area of an active element is proportional to its loading, resulting in a linear relation
between catalyst activity and loading. However, nonlinear dependence of surface
area on loading can exist, especially at high catalyst concentrations. In this case,
the optimization of the catalyst activity distribution should take this dependence
into account. For structure-sensitive reactions, the relation between reaction rate
and surface area depends strongly on the catalytic system, so that the optimal
distribution problem has to be dealt with case by case.
2.4.1 Factors Affecting Catalyst Dispersion

The catalytic surface area of a supported catalyst (or equivalently its dispersion)
depends significantly on its preparation procedure. A common technique for this
is impregnation of the support with a precursor solution. The conditions of the
impregnation step, as well as the subsequent steps of drying, calcination, and re-
duction, affect dispersion and hence the dependence of catalytic surface area on
catalyst loading. More specifically, parameters that influence catalyst dispersion
include the nature and concentration of precursor and support, the duration, tem-
perature, and atmosphere (in contact with the catalyst) of the preparation steps,
etc. (Delmon and Houalla, 1979). The effects of these parameters on catalyst
properties are complex, and have been the topic of many studies over the years
(cf. Schwarz et al., 1995). In the following, only a few representative cases are
discussed, in order to demonstrate underlying phenomena affecting the active
surface area of a supported catalyst.
Impregnation
During impregnation, the active agent is introduced into the support as a so-
lution of a precursor compound, which is typically not its final form. Catalyst
dispersion is affected in this step by the interaction of precursor and support (e.g.
strong or weak adsorption). This in turn depends on the precursor properties and
concentration, nature of the support, solution pH, ionic strength, and presence
of added or extraneous ions. The effects of these parameters on adsorption are

discussed in more detail in Chapter 7. Generally, when strong interaction exists
during impregnation, the precursor establishes a physical or chemical bond with
the support surface. These interactions result in a near-atomic distribution of the
precursor on the support, and thus in high dispersion.
When only weak precursor–support interaction exists, the support acts merely
as a physical surface. In this case, the catalyst dispersion is typically lower than in
the case of strong interaction. Metal particles develop through crystallization of
precursor in the pore-filling solution during solvent evaporation. Crystallite sizes
are dictated by mass transfer during precipitation–crystallization of the dissolved
component, which in turn is controlled by the conditions during solvent evapo-
ration (Le Page, 1987). Weak precursor–support interaction can be encountered
in the following cases: (a) the support is inert, (b) the precursor ion has the same
charge as the support surface, (c) the support is in a range of pH where it is not
charged, (d) the adsorption sites of the support are occupied.
The surface of α-Al2 O3 is inert because it is dehydroxylated and hence cannot
develop surface charge. Hence, catalysts based on α-Al2 O3 have typically low
dispersion (cf. Gavriilidis et al., 1993b). Interaction between support and catalyst
precursor can be affected by the combination of precursor and support used.
When the support surface is negatively charged in solution, as is the case with silica
(Brunelle, 1978), the metal will not be adsorbed if it is contained in the anion, as in
H2 PtCl6 , but will be adsorbed if it is present in the cation, as in [Pt(NH3 )4 ]Cl2 . No
adsorption occurs even when [Pt(NH3 )4 ]Cl2 is used as the precursor if the pH is
below 6 (Benesi et al., 1968). Silica has a zeta-potential value of almost zero for pH
0–6, indicating that in this pH range the charge of the surface is practically zero,
and hence it behaves as an inert support (Anderson, 1975; Brunelle 1978). The
above picture of the support–precursor interaction is simplistic, but it provides
a framework within which the behavior of a large number of support–precursor
systems can be understood.
Finally, every support material has a finite adsorption capacity. If the amount of
metal ion to be deposited is larger than the adsorption capacity at a given pH, then
the excess amount will deposit without any interaction with the support surface
(Che and Bonneviot, 1988). For example, for iridium–titania systems, it has been
observed that two types of particles are formed: small particles produced from
precursor ion-exchanged on the support surface, and large particles produced by
precursor not interacting with the support surface (Van Tiep et al., 1986).
Drying
The drying process can affect metal particle size, primarily when weak support–
precursor interaction exists. In this case, catalyst particles are formed by crystal-
lization of the precursor from the pore-filling liquid during solvent evaporation.
The ratio of nucleation rate to crystal growth rate determines the crystal size; if
it is large, the crystal size is small (Le Page, 1987). A high internal surface area
favors heterogeneous nucleation. The nucleation–crystallization process is influ-
enced by mass and heat transfer in the porous structure. When evaporation of the
solvent is fast compared to solute diffusion, the precursor concentration increases

at the air–liquid interface, which gradually recedes to the pellet interior. Hence,
particles deposit at the interface, leading to relatively homogeneous deposition
and particle size throughout the pellet.
If the evaporation is slow, the particle size depends on the pore structure as well
as on the precursor concentration of the impregnating solution. Solvent removal
concentrates the solution to the point where precursor crystallization begins, pro-
viding nuclei in the pores still containing solution. The crystallites will grow as
long as there is precursor-containing solution surrounding them. Large pores can
accommodate large solution volume, and hence crystallite size is related to pore
size. Indeed, it has been observed that supports with large pore size give rise to
large catalyst particles (Dorling et al., 1971; Hamada et al., 1987). In addition, the
pore structure can affect the availability of precursor for crystallite growth, by
leading to formation of isolated clusters of liquid-filled pores.
Similar arguments can be used to explain the effect of precursor concentration
on particle size (Dorling and Moss, 1967; Dorling et al., 1971; Moss et al., 1979). If
the impregnating solution is dilute and solute diffusion is fast, many pores empty
before crystallization begins. Increasing the concentration of the impregnating
solution increases the number of nuclei, and at a certain precursor concentration,
crystallization starts in practically all pores. Up to this point, the mean crystallite
size is almost constant, and the catalyst surface area is proportional to the catalyst
loading. At higher precursor concentration, though, the number of crystallites
remains constant but the crystallite size increases, and hence the catalyst surface
area does not increase linearly with further catalyst loading.
Calcination
The calcination step, which is usually carried out in oxidizing atmosphere, can lead
to a number of transformations (Che and Bennett, 1989): precursor decomposition
and formation of an oxide species, bonding of the formed oxide with the support,
removal of some of the elements introduced during the preparation, decomposi-
tion of the precursor ionic complex, ligand exchange reactions between surface
anchoring groups and ligands bound to the metal ion, elimination of carbonaceous
impurities, and sintering of the precursor compound or the formed oxide species.
For platinum–silica catalysts, crystallite size was found to increase with calci-
nation temperature (Dorling et al., 1971; Brunelle et al., 1976). When catalysts
were obtained by chloroplatinic acid impregnation (no support–precursor inter-
action), the increase was more substantial than with catalysts prepared by amine
complex (interaction through ion exchange). Furthermore, dispersion loss started
at lower temperature for the former type of catalysts. For Pt/γ -Al2 O3 catalysts,
Bournonville et al. (1983) found that the metal dispersion exhibited a maximum as
a function of calcination temperature. It was proposed that increasing calcination
temperature facilitated the formation of an oxochlorinated platinum complex,
which starts to decompose above a certain temperature. The formation of this
complex was crucial for obtaining high dispersion. This complex was strongly
bound to the carrier and led to the formation of a well-dispersed metallic phase
upon reduction. Catalysts that were reduced without previous calcination had
significantly lower dispersion than those subjected to calcination before reduc-

tion, as also found by Carballo et al. (1978). Dispersion loss at high calcination
temperature was more pronounced at higher platinum loadings.
High-temperature treatments such as calcination and reduction can lead to
sintering of the precursor compound or the formed oxide species. This results in
larger particles and loss of catalyst dispersion. Excellent reviews of sintering phe-
nomena have appeared in the literature (Wanke and Flynn, 1975; Ruckenstein
and Dadyburjor, 1983; Bartholomew, 1993). High-temperature treatments can
also lead to increase of catalyst dispersion. It has been demonstrated that oxida-
tion treatments or oxidation–reduction cycles at high temperature can result in
redispersion of large metal particles (cf. Ruckenstein and Malhotra, 1976; Wang
and Schmidt, 1981). This has been attributed to the lower surface tension of the
metal oxide than of the metal, causing a flattened structure to be formed from a
hemispherical particle of the reduced metal, which is readily reconstructed into
smaller hemispherical particles upon reduction of the oxidized structures (Lee
and Ruckenstein, 1983).
Reduction
Reduction in hydrogen is commonly employed for catalyst activation, during
which the metal precursor compound or its oxide is transformed into the metallic
state (Foger, 1984). The interaction of the oxide with the support, which depends
on the phenomena that have taken place during the preceding calcination step,
can also affect the reduction step (Delmon and Houalla, 1979). Various effects of
reduction temperature on particle size have been observed.
Increasing the temperature of reduction was found to result in an increase in
particle size for Pt/γ -Al2 O3 catalysts prepared by chloroplatinic acid impregnation
(Wilson and Hall, 1970). Only a small effect of reduction temperature on platinum
particle size was observed for catalysts prepared by a platinum amine complex ad-
sorption and reduced at 400–700◦ C (Brunelle et al., 1976). For Ni/Al2 O3 catalysts,
the nickel surface area showed a rather weak dependence on reduction temper-
ature between 300 and 500◦ C, with a maximum occurring at about 350–400◦ C.
Substantial loss of nickel surface area was obtained if the samples were calcined
before reduction (Bartholomew and Farrauto, 1976). Other treatments before re-
duction can also affect catalyst dispersion. Sarkany and Gonzalez (1983) showed
that initial pretreatment in helium results in Pt/γ -Al2 O3 catalyst with dispersion
considerably larger than that obtained after reduction. In this case, the dispersion
reached a value of 60% at loading of 2 wt% and remained constant up to 6 wt%
after helium pretreatment, whereas after reduction it decreased from 50% to 20%
as the loading increased from 2 to 6 wt%.
2.4.2 Dependence of Catalytic Surface Area on Catalyst Loading

From the above discussion, it is clear that catalytic surface area depends strongly
on the preparation conditions. A variety of experimental techniques are avail-
able to measure catalytic surface area (or equivalently catalyst dispersion) and
catalyst loading (cf. Delannay, 1984). Such measurements have been performed
for various catalysts such as Pt/SiO2 (Dorling et al., 1971; Brunelle et al., 1976),
Pt/Al2 O3 (Basset et al., 1975), Rh/SiO2 (Arakawa et al., 1984; Underwood and
Bell, 1987), Rh/Al2 O3 (Fuentes and Figueras, 1980), Pd/SiO2 (Moss et al., 1979),
Pd/Al2 O3 (Scholten and Van Montfoort, 1962; Aben, 1968), Ag/SiO2 (Seyed-
monir et al., 1990), Ag/Al2 O3 (Gavriilidis et al., 1993b), Ni/Al2 O3 (Bartholomew
et al., 1980; Huang and Schwarz, 1987), Co/Al2 O3 (Reuel and Bartholomew, 1984),
Ir/Al2 O3 (Barbier and Marecot, 1981), Ru/Al2 O3 (King, 1978), and V2 O5 /ZrO2
(Chary et al., 1991).
A relationship between active element surface area per unit weight of catalyst
pellet, A, and weight fraction of active catalyst, q, which can be used to represent
many real situations is
pq
A= (2.122)
1 + bq
where p represents the specific surface area of active catalyst (square meters per
gram of active catalyst) and p/b is surface area of active catalyst per unit weight of
catalyst pellet at saturation, i.e. at high loadings. Note that from the active element
surface area A, the catalyst dispersion D, defined as the fraction of active element
atoms available on crystallite surface, is obtained as
D 1
= (2.123)
D0 1 + bq
where D0 is the dispersion as the loading q → 0. Dorling et al. (1971) prepared

Pt/SiO2 catalysts by impregnation of silica with chloroplatinic acid. In agreement
with discussion in section 2.4.1, since chloroplatinic acid does not interact with
silica, a nonlinear dependence of surface area on loading was found as shown
graphically in Figure 2.18. On the other hand, when an amine platinum complex
which interacts with silica was used, a linear dependence was found for catalysts
containing up to 4.5 wt% Pt.
Basset et al. (1975) and Szegner et al. (1997) used chloroplatinic acid for
γ -Al2 O3 impregnation, and found a linear dependence of catalyst surface area
Figure 2.18. Variation of platinum sur-

face area with platinum content for
Pt/SiO2 catalysts. The catalysts were pre-
pared by impregnation with H2 PtCl6
and [Pt(NH3 )4 ]Cl2 . Solid line is equation
(2.122) with p = 84 m2 /g, b = 10. (From
Dorling et al., 1971.)
Figure 2.19. Variation of platinum

surface area with platinum content
for Pt/γ -Al2 O3 catalysts. The cata-
lysts were prepared by impregna-
tion with H2 PtCl6 and Pt(NO2 )2
(NH3 )2 . Solid line is equation
(2.122) with p = 109 m2 /g, b = 58.
(From Basset et al., 1975.)
on loading, because in this case strong precursor–support interactions exist. Both

researchers observed a linear dependence up to 3 wt% Pt as shown in Figure 2.19.
Beyond this, the platinum surface area remained constant, at ∼5 m2 /g, presumably
due to full occupation of the adsorption sites. Further, Szegner et al. (1997) found
that addition of tin to these catalysts resulted in no significant change in the ac-
cessible platinum atoms or dispersion. Basset et al. (1975) also prepared catalysts
using Pt(NO)2 (NH3 )2 , and these showed a nonlinear relationship between sur-
face area and loading as demonstrated in Figure 2.19. For the catalysts described
above, the plot of active surface area vs catalyst loading follows equation (2.122),
as shown by the corresponding solid curves in Figures 2.18 and 2.19.
Similar behavior is also exhibited by other catalytic systems. For example, Moss
et al. (1979) prepared palladium catalysts by impregnating silica with tetraam-
minepalladous nitrate. The metal surface area measured by CO chemisorption is
shown in Figure 2.20. Results for Ag/α-Al2 O3 catalysts prepared by silver lactate
Figure 2.20. Variation of palladium

surface area with palladium content
for Pd/SiO2 catalysts. The catalysts
were prepared by impregnation with
Pd(NH3 )4 (NO3 )2 . Solid line is equa-
tion (2.122) with p = 177 m2 /g, b =
22. (From Moss et al., 1979.)
Figure 2.21. Variation of silver sur-

face area with silver content for
Ag/α-Al2 O3 catalysts. The cata-
lysts were prepared by impreg-
nation with silver lactate. Solid
line is equation (2.122) with p =
0.66 m2 /g, b = 1.08. (From Gavri-
ilidis et al., 1993b.)
impregnation (Gavriilidis et al., 1993b) are shown in Figure 2.21. The nonlinear
dependence of surface area on loading is due to the fact that α-Al2 O3 is an inert
support. In Figure 2.22, catalyst dispersion measured by hydrogen chemisorption
as a function of loading is shown for Rh supported on SiO2 . The catalysts were
prepared by incipient wetness impregnation of SiO2 , using aqueous solutions of
Rh(NO3 )3 (Underwood and Bell, 1987).
2.5 Optimal Distribution of Catalyst Loading

Since catalyst loading is the quantity controlled during catalyst preparation, it is
preferable to use as the optimization variable. The focus is on reactions which
are essentially structure-insensitive, i.e. for which we can assume that the rate of
reaction r depends linearly on the active element surface area A:
r = r A (2.124)
where r is the reaction rate per unit weight of the catalyst pellet, and r is the
specific reaction rate per unit surface area of active element. By using equation
Figure 2.22. Variation of rhodium

dispersion with rhodium content for
Rh/SiO2 catalyst. The catalysts were
prepared by incipient-wetness im-
pregnation with Rh(NO3 )3 . Solid
line is equation (2.123) with b = 42.
(From Underwood and Bell, 1987.)
(2.122) the relation between reaction rate and catalyst loading is obtained:
pq
r = r (2.125)
1 + bq
2.5.1 The Problem Formulation

With a quantitative relation between reaction rate r and catalyst concentration q
available, the optimization problem can now be reformulated (Baratti et al., 1993).
The aim is to identify the concentration profile q(s) of active element within the
support which, for a fixed total amount, maximizes a given performance index of
the catalyst pellet.

The optimization variable in dimensionless form is
q(s)
µ(s) = . (2.126)
q̄
Since q̄ is the volume-average value of q(s), µ(s) must satisfy the condition
1
(n + 1) µ(s)s n ds = 1. (2.127)
0
In addition, on physical grounds it is reasonable to assume that µ(s) is positive

and bounded, i.e.,
0 ≤ µ(s) < α (2.128)
where α is the maximum allowable value of the local active element concentration
on the support.
The reaction rate can be represented using equations (2.125) and (2.126) as
follows:
µ(s)
r = r0 . (2.129)
1 + Bµ(s)
The factor r0 represents the rate of reaction in a catalyst pellet with maximum
dispersion, i.e. where the active element surface area per unit catalyst weight is
equal to pq̄. The dimensionless quantity B = bq̄ is a measure of the nonlinearity
of the A-vs-q relation; B = 0 corresponds to a linear relation, while for increasing
B the nonlinearity becomes more severe. For a fixed catalyst preparation method
(i.e. fixed b value), B increases with the average active element loading of the
pellet, q̄.
As in section 2.3.2, we consider a general reacting system, where any number
of reactions occur, following arbitrary kinetics, in the presence of external mass
and heat transport resistances. The mass and energy balances in matrix form are

1 d dx µ(s)
L[x] = n sn = w(x) (2.130)
s ds ds 1 + Bµ(s)
with BCs
dx
s = 0: =0 (2.131a)
ds
dx
s = 1: = A (xf − x) (2.131b)
ds
where the vectors x, xf , and w and matrix A are defined in (2.101) and (2.102).
The effectiveness factor η M is defined as the ratio between the actual rate of con-
sumption of reactant M and the corresponding rate in the absence of all transport
resistances in a catalyst pellet with the same overall loading of active element
distributed uniformly throughout the support [i.e. µ(s) = 1]. Note that the pellet
used as reference in this definition is uniform while that under examination is
nonuniform, so they may have different values of the overall active surface area.
Thus
1
(n + 1)(1 + B) µ(s)
ηM = J w M (x)s n ds. (2.132)
DM j=1 ν Mj φ j 0
2 1 + Bµ(s)
Similarly, the selectivity and yield are defined by

1 µ(s)
w N (x)s n ds
DM 1 + Bµ(s)
0
SNM = (2.133)
DN 1 µ(s)
w M (x)s n ds
0
1 + Bµ(s)
and

(n + 1)(1 + B) 1
µ(s)
YNM = η M SNM = w N (x)s n ds. (2.134)
DN Jj=1 ν Mj φ 2j 0 1 + Bµ(s)
Thus, summarizing, the optimization problem consists in evaluating the func-

tion µ(s), which maximizes one of the performance indexes above, under the
constraints of the diffusion–reaction equations (2.130)–(2.131) and of equations
(2.127) and (2.128).
Development of the Hamiltonian

The optimization procedure is based on the variational method. As described
in Appendix A, this requires derivation of the relevant Hamiltonian which is a
function providing the overall effect of changes in the optimization function. For
this, it is convenient to rewrite the original problem (2.130) by introducing the
new variables γ and h as follows:
dx
= γs −n (2.135)
ds
dγ µ(s)
= w(x)s n (2.136)
ds 1 + Bµ(s)
dh
= µ(s)s n (2.137)
ds
where the last equation arises from the integral constraint (2.127). The corre-
sponding BCs follow from equations (2.127), (2.131), and (2.135):
s = 0: γ=0 (2.138a)
s = 1: γ = A(xf − x) (2.138b)
s = 0: h=0 (2.138c)
1
s = 1: h= . (2.138d)
n+1
When the effectiveness factor (2.132) is the objective functional, the Hamiltonian
is defined by (cf. Appendix A)
µ(s)
H= w M (x) + λT2 w(x) s n + λT1 γs −n + λ3 µ(s)s n (2.139)
1 + Bµ(s)
where λ1 , λ2 (column vectors, with I + 1 elements), and λ3 are the Lagrange
multipliers defined by the following differential equations:
dλ1 µ(s)
=− s n [w Mx + JT λ2 ] (2.140)
ds 1 + Bµ(s)
dλ2
= −λ1 s −n (2.141)
ds
dλ3
=0 (2.142)
ds
where w Mx is the column vector of derivatives of w M with respect to x, and J is
the Jacobian matrix of vector w with respect to the independent variable x:
 
∂w M /∂ x1
 ∂w M /∂ x2 
∂w M   ..


w Mx = = .  (2.143)
∂x  
 ∂w M /∂ x1 
∂w M /∂ xI+1
 
∂w1 /∂ x1 ∂w1 /∂ x2 . . . ∂w1 /∂ xI ∂w1 /∂ xI+1
 ∂w2 /∂ x1 ∂w2 /∂ x2 . . . ∂w2 /∂ xI ∂w2 /∂ xI+1 
 
 .. .. .. .. 
J= . . . . . (2.144)
 
 ∂w I /∂ x1 ∂w I /∂ x2 . . . ∂w I /∂ xI ∂w I /∂ xI+1 
∂w I+1 /∂ x1 ∂w I+1 /∂ x2 . . . ∂w I+1 /∂ xI ∂w I+1 /∂ xI+1
The relevant BCs are
s = 0: λ1 = 0 (2.145a)
s = 1: λ1 = AT λ2 . (2.145b)
When the yield YNM is considered as the objective function, the same expres-
sions for the Hamiltonian and the Lagrange multipliers are obtained, but w M (x)
is now replaced by w N (x) (cf. Appendix A).
In the case where the objective functional is the selectivity SNM , it is convenient
to introduce two auxiliary variables xN and xM defined as follows:
dxN µ(s)
= w N (x)s n (2.146)
ds 1 + Bµ(s)
dxM µ(s)
= w M (x)s n (2.147)
ds 1 + Bµ(s)
with initial conditions
s = 0: xN = 0 (2.148a)
s = 0: xM = 0 (2.148b)
Using the above equations, the selectivity definition (2.133) becomes

DM xN (1)
SNM = (2.149)
DN xM (1)
The variational problem now consists of the objective functional (2.149), where
the state variables satisfy the differential equations (2.135)–(2.137), (2.146), and
(2.147), with BCs (2.138) and (2.148). The relevant Hamiltonian is (cf. Appendix A)
µ(s) T
H= λ2 w(x) + λ4 w N (x) + λ5 w M (x) s n + λT1 γs −n + λ3 µ(s)s n
1 + Bµ(s)
(2.150)
where the Lagrange multipliers are defined by the differential equations (2.141)
and (2.142) together with
dλ1 µ(s)
=− s n (JT λ2 + λ4 w Nx + λ5 w Mx ) (2.151)
ds 1 + Bµ(s)
dλ4
=0 (2.152)
ds
dλ5
=0 (2.153)
ds
along with BCs (2.145) and
DM
s = 1: λ4 = (2.154)
DN xM
DM xN
s = 1: λ5 = − 2
. (2.155)
DN xM
As explained in Appendix A, the solution of the optimization problem is given
by the active element concentration µ(s) which maximizes the Hamiltonian H
defined by equation (2.139) or (2.150), depending on the specific performance
index considered as the objective functional, while simultaneously allowing for
the constraint (2.128). The solution is described below for four cases (Baratti et al.,
1993, 1997). The first is the case of a single first-order isothermal reaction where
an analytical solution can be obtained, the second the case of linear dependence
between the active element surface area and its concentration, and the third the
case of a single first-order nonisothermal reaction where a numerical method of
general applicability is used for the solution. In the fourth case, the same method is
applied to investigate complex reacting systems which give rise to unusual optimal
loading distributions.
2.5.2 A Single First-Order Isothermal Reaction

For the case of a single first-order isothermal reaction, the system of differential
equations (2.130)–(2.131) reduces to
µ(s)
L[u] = φ 2 u (2.156)
1 + Bµ(s)
with BCs
du
s = 0: =0 (2.157a)
ds
s = 1: u=1 (2.157b)
where interphase mass transfer resistance has been neglected. Following the gen-
eral procedure outlined above, the original problem (2.156) can be rewritten as
follows:
du
= γ s −n (2.158)
ds
dγ µ(s)
= φ2 s nu (2.159)
ds 1 + Bµ(s)
dh
= µ(s)s n (2.160)
ds
with the corresponding conditions
s = 0: γ =0 (2.161a)
s = 1: u=1 (2.161b)
s = 0: h=0 (2.161c)
1
s = 1: h= . (2.161d)
n+1
Using the effectiveness factor as the objective functional, the Hamiltonian is given
by (2.139):
µ(s)
H= (1 + λ2 )φ 2 us n + λ1 γ s −n + λ3 µ(s)s n (2.162)
1 + Bµ(s)
and the Lagrange multipliers by (2.140)–(2.142):

dλ1 µ(s)
=− (1 + λ2 )φ 2 s n (2.163)
ds 1 + Bµ(s)
dλ2
= −λ1 s −n (2.164)
ds
dλ3
=0 (2.165)
ds
with BCs
s = 0: λ1 = 0 (2.166a)
s = 1: λ2 = 0. (2.166b)
According to the maximum principle, in the case where the optimization function
is not bounded [i.e., the constraint (2.128) is not included, or α = ∞], the necessary
condition for the Hamiltonian H to be maximum leads to
∂H 1
= (1 + λ2 )φ 2 us n + λ3 s n = 0. (2.167)
∂µ [1 + Bµ(s)]2
By comparing equations (2.158) and (2.159) with equations (2.163) and (2.164),
together with the corresponding BCs (2.161a), (2.161b) and (2.166a), (2.166b), we
obtain
u = 1 + λ2 (2.168)
γ = −λ1 (2.169)
which may be verified by direct substitution. Using equation (2.168), the optimality
condition (2.167) reduces to
√
u −λ3
= (2.170)
1 + Bµ(s) φ
which, substituted in equations (2.158)–(2.160), leads to a system of three ODEs
with four BCs (2.161a–d). This set can be solved analytically to yield the three
functions u(s), γ (s), and h(s) in addition to λ3 , which according to equation (2.165)
is constant. The analytical expressions for the optimal active element distribution
µ(s) , the reactant concentration profile u(s), and the catalyst pellet effectiveness
factor η are reported in Table 2.4 for three pellet geometries: slab (n = 0), cylinder
(n = 1), and sphere (n = 2).
In Figure 2.23.a, the optimal distribution of the active element µ(s) is shown √
for a sphere and for various values of the dimensionless parameter ψ = φ/ B,
which can be regarded as a modified Thiele modulus that takes into account the
nonlinearity in the dependence of the active surface area upon the active element
concentration. Note that this is the only variable needed to fully characterize the
optimal distribution. For values of ψ approaching 0, the optimal distribution ap-
proaches the uniform distribution. This is physically sound, considering that when
the concentration of the reactant inside the pellet is uniform (small φ values) or
Table 2.4. Optimal active element distribution, with the corresponding reactant
concentration profile and the effectiveness factor,
√for three pellet geometries and for
a single first-order isothermal reaction (ψ = φ/ B).
Active element distribution Reactant concentration Effectiveness factor

n µ(s) u(s) η = (n + 1)(1 + B)τ
cosh ψs cosh ψs tanh ψ
0 ψ (1 − τ ) +τ τ=
sinh ψ cosh ψ tanh ψ + ψ B
ψ I0 (ψs) I0 (ψs) I1 (ψ)
1 (1 − τ ) +τ τ=
2 I1 (ψ) I0 (ψ) Bψ
I
2 0
(ψ) + I1 (ψ)
ψ2 sinh ψs 1 − τ sinh ψs ψ − tanh ψ
2 +τ τ=
3s ψ cosh ψ − sinh ψ s sinh ψ ( Bψ3
2
− 1) tanh ψ + ψ
when the nonlinearity in the A-vs-q dependence is strong (large B values), the cat-
alyst is more effectively utilized when it is distributed as uniformly as possible. On
the other hand, when ψ → ∞, the optimal distribution becomes a progressively
steeper monotonically increasing function of s, exhibiting its maximum value at
the pellet external surface. This occurs when either B → 0 or φ → ∞. For B → 0,
since there is no problem of loss of catalyst surface area with increasing loading,
it is better to locate all the catalyst where the reactant concentration is maximum,
i.e. at the external surface of the pellet. For increasing values of φ, no matter how
large B is, the reaction takes place in a narrower region near the pellets external
surface, which is the only place where the active element can be profitably located.
Figure 2.23. (a) Optimal active element distribution for various values of
the modified Thiele modulus ψ, and (b) corresponding values of the effec-
tiveness factor. Reaction system: a single isothermal first-order reaction
in a sphere (n = 2). (From Baratti et al., 1993.)
This can be seen explicitly from the expressions for the optimal µ(s) reported in
Table 2.4, where for any of the examined geometries, in the limit as ψ → ∞,
ψ
µ(s) → exp[−ψ(1 − s)]. (2.171)
(n + 1)s n/2
Hence, in the limit of ψ → ∞, µ(s) approaches a Dirac-delta function located at

the pellet’s external surface. This result is consistent with all the previous cases
examined, where a linear A-vs-q dependence was implicitly assumed.
The corresponding optimal values of the effectiveness factor η are shown in
Figure 2.23.b as a function of the Thiele modulus φ, for various values of B.
For B = 0, the optimal catalyst distribution is a Dirac-delta function located at
the pellet external surface. In this case, the effectiveness factor is 1 regardless of
the value of the Thiele modulus. However, for increasing B values, η decreases,
because the catalyst has to be spread out towards the pellet center for optimal
utilization. The decrease is more significant at large Thiele modulus. Indeed, for
finite B and φ → ∞ , using the relationships reported in Table 2.4 it can be shown
that
(n + 1)(1 + B)
η→ √ (2.172)
φ B
so that the η-vs-φ dependence is similar to that for uniform pellets (cf. Aris, 1975).
2.5.3 Linear Dependence between the Active Element Surface Area and Its Loading
When the dependence between the active element surface area and its loading
is linear, i.e. B = 0, the expressions of both Hamiltonians (2.139) and (2.150) are
linear with respect to µ, and can be reduced to the following form:
H = µ + ω (2.173)
The function is called switching function, and its expression depends on the reac-
tion system under consideration as well as the performance index to be optimized.
Both and ω are independent of µ. In particular, considering the performance
indices (2.132)–(2.134), the following expressions for the switching function are
obtained:

= w M (x) + λT2 w(x) + λ3 s n for η M (2.174)

= w N (x) + λT2 w(x) + λ3 s n for YNM (2.175)
T n
= λ2 w(x) + λ3 + λ4 w N (x) + λ5 w M (x) s for SNM (2.176)
In optimal-control theory, a system for which the Hamiltonian is a linear function

of the control variable (which in this case corresponds to µ), and the latter is
constrained between a minimum and a maximum setting [which corresponds to
equation (2.128)], is known as a bang–bang control system (cf. Ray, 1981). In this
case, the Hamiltonian is maximized by setting the control variable at its maximum
value if the switching function is positive and at its minimum value if the switching
function is negative. Hence, in general, the optimal distribution µ(s) is given by a
multiple step function, where the positions and number of steps are controlled by
the number of sign changes of the switching function . Accordingly, the optimal
active element concentration profile µ is given by

α when > 0
µ= (2.177)
0 when < 0.
For example, in the simple case of a single first-order isothermal reaction with
the effectiveness factor as the objective functional treated in section 2.5.2, the
switching function can be readily obtained from equations (2.162) and (2.168):
= (u2 φ 2 + λ3 )s n . (2.178)
Since u2 and s n are increasing functions of s and λ3 is constant [see equation

(2.165)], is a monotonically increasing function and has only one sign change
for s ∈ [0, 1]. This implies that the optimal distribution is given by a single step
of height α, located at the pellet surface (i.e. an eggshell distribution), whose
half-width satisfies the following relation derived from constraint (2.127):
1
1 − (1 − 2)n+1 = . (2.179)
α
It can be readily seen that in the case where the optimal distribution is not bounded
from above, i.e. α → ∞, then → 0 and the distribution approaches a Dirac-delta
located at s = 1. Thus, the Dirac-delta is in fact a special case (for α → ∞) of the
optimal distribution, while the step distribution is optimal for finite α.
For other reacting systems, the expression for the switching function , whose
sign changes dictate the number of steps of the optimal distribution, will be dif-
ferent. Insight can be gained by studying the switching function, realizing that for
complex reaction systems this study can become cumbersome. In this case, the
numerical method presented in the next section may provide a more convenient
alternative.
2.5.4 First-Order Nonisothermal Reactions: Numerical Optimization

In sections 2.5.2 and 2.5.3, two situations were examined where the optimal cat-
alyst distribution was obtained analytically. However, in many cases it is more
efficient to solve the optimization problem directly through a suitable numerical
technique, which is presented next. Using the orthogonal collocation method (cf.
Finlayson, 1980; Villadsen and Michelsen, 1978), the diffusion–reaction differen-
tial equations reduce to a set of nonlinear algebraic equations, whose solution is
obtained through the Newton–Raphson method. The values of the active element
concentration at the collocation points are regarded as the adjustable parameters
of the optimization problem. The performance index of interest is evaluated using

Gauss–Jacobi quadrature formulae, where the quadrature points coincide with
the collocation points (Villadsen and Michelsen, 1978).
The optimal catalyst distribution can be a discontinuous function. For example,
in section 2.5.3 (where B = 0) it is a step function. Based on continuity argu-
ments, it is reasonable to expect that for nonzero but small B values the optimal
distribution will be different from a step distribution but retain the feature of
being nonzero only in a small interval of the domain 0 ≤ s ≤ 1 and zero outside,
while exhibiting discontinuities at the interval boundaries. In such a case, the op-
timal solution can be obtained conveniently by using orthogonal collocation on
finite elements, where the boundaries are located where the discontinuities arise.
However, such locations are not known a priori, and therefore the numerical de-
termination of the optimal distribution becomes difficult. For this purpose, during
optimization the elements are not fixed but have moving boundaries located at
the discontinuity points.
Specifically, the optimization is carried out as follows:
1. The optimization problem is solved using orthogonal collocation applied

to the entire domain 0 ≤ s ≤ 1 with an increasing number of discretization
points (typically 4 to 7).
2. If the obtained solution exhibits regions where µ = 0, then orthogonal
collocation on finite elements is applied, by placing the element boundaries
so that each region where µ = 0 constitutes a single element. In order to
allow for discontinuities, µ is allowed to take two different values at the
element boundaries, depending on which element it belongs to.
3. The above procedure is iterated, by moving the location of the element
boundaries until in each element the optimal solution µ is either always
zero or always nonzero.
4. Convergence of the procedure is checked by verifying that the calculated
optimal distribution does not change on increasing the number of colloca-
tion points in each element.
A single first-order reaction

The optimal catalyst distributions computed for a single first-order nonisothermal
reaction and B ≤ 0.1 are shown in Figure 2.24. Note that these values of B imply
volume-average catalyst loadings less than 1 wt% for the Pt/SiO2 system in Figure
2.18, 0.5 wt% for the Pd/SiO2 system in Figure 2.20, and 0.25 wt% for the Rh/SiO2
system in Figure 2.22, and thus cover the range of interest in most practical ap-
plications. For low B values, the optimal distribution exhibits some discontinuity
points and strongly resembles a step distribution. For increasing values of B, high
local catalyst concentration would result in significant loss of active surface area.
This provides a motivation for spreading the active element. However, the latter
is counterbalanced by the effect of the reactant concentration and temperature
gradients, which require all the catalyst to be located where the reaction rate is
Figure 2.24. Optimal active element

distribution for α = 7 and various val-
ues of B. Reaction system: a single
nonisothermal, first-order reaction,
with β = 0.2, γ = 18, φ = 1, Bim = ∞,
Bih = ∞, and n = 2. (From Baratti
et al., 1993.)
maximum. As a consequence, the values of the effectiveness factor corresponding

to the optimal active element distribution decrease for increasing values of B,
as shown in Figure 2.25. For comparison, the values of the effectiveness factor
corresponding to the uniform distribution are also shown. These conclusions are
expected to be of general validity, and in fact have also been reached for the case
of bimolecular Langmuir kinetics (Baratti et al., 1993).
Figure 2.25. Optimal values of the effectiveness factor as

a function of B. Reaction system: a single nonisothermal,
first-order reaction, with β = 0.2, γ = 18, φ = 5, Bim = ∞,
Bih = ∞, and n = 2. (From Baratti et al., 1993.)
Figure 2.26. Ratio of effectiveness factors for optimal and uniform distributions. Reaction
system: a single, nonisothermal, first-order reaction, with γ = 18, Bim = ∞, Bih = ∞, and
n = 2. (From Baratti et al., 1993.)
The effectiveness factor ratio, ηoptimal /ηuniform , is a measure of the improve-

ment that can be obtained by the optimal catalyst distribution as compared to the
uniform catalyst. This ratio is shown for various selected combinations of β, α,
and φ values in Figure 2.26. It is evident that the optimal distribution provides a
substantial improvement in the catalyst performance, as compared to the uniform
distribution. The extent of improvement increases as β, α, or φ increases, and is
larger for smaller B values.
Two consecutive first-order reactions

The case of two nonisothermal consecutive first-order reactions is shown in Figure
2.27. It may be seen that the optimal distribution for selectivity maximization
remains a step distribution even for large values of B, while the optimal distribution
for effectiveness factor maximization spreads out with increasing B. Such behavior
can be justified by considering that in the case of selectivity optimization, there
(a) (b)
Figure 2.27. Optimal active-element distributions for α = 7 and various values

of B, in the case of (a) selectivity and (b) effectiveness factor optimization.
Reaction system: two nonisothermal, first-order, consecutive reactions, with
β1 = 0.2, β2 = 0.1, γ1 = 20, γ2 = 10, φ = 1, D1 = D2 = 1, Bim1 = Bim2 = ∞,
Bih = ∞, and n = 2. (From Baratti et al., 1993.)
is no issue of performance loss owing to poor catalyst utilization in terms of the

active surface area; the key parameter is the ratio of the two reaction rates. This
is not the case for performance indices based on catalyst productivity, which lead
to broader optimal distributions with increasing values of B. Thus, we have the
important conclusion that even for the same reaction network and physicochemical
parameters, the optimal catalyst distribution depends on the choice of the catalyst
performance index selected for optimization.
2.5.5 Multistep Optimal Loading Distribution

Even in the simple case where the catalyst surface area depends linearly on loading
(section 2.5.3), the theoretical result indicated that the optimal catalyst distribution
can be a multiple-step function. This was demonstrated recently for the case of
two parallel, irreversible reactions (Baratti et al., 1997)
1
A −→ B
2
(2.180)
A −→ C
whose dimensionless rates are given by the Langmuir form
um j exp[γ j (1 − 1/θ )]
f j (u, θ) = , j = 1, 2. (2.181)
{1 + σ j u exp[ε j (1 − 1/θ )]}2
In the case of negligible external mass and heat transfer resistances and equal
heats of reaction, Prater’s relationship between the dimensionless reactant con-
centration u and dimensionless temperature θ leads to [see also equation (2.43)]
θ = 1 + β(1 − u). (2.182)
This relation allows one to represent the overall dimensionless rate of consumption
of reactant A as a function of only one variable:
k 2,f
f (u) = f1 [u, 1 + β(1 − u)] + f2 [u, 1 + β(1 − u)]. (2.183)
k 1,f
The behavior of the global reaction rate f (u) is shown in Figure 2.28 for three
values of k2,f /k1,f , which is the ratio of the reaction constants at bulk fluid temper-
ature. These values were chosen for illustrative purposes, so as to clearly separate
the two maxima of function f. It is assumed that the dependence of catalyst activ-
ity on loading is linear (i.e. B = 0). We seek to maximize the overall effectiveness
factor

n+1 1
η= µ(s) f (u)s n ds (2.184)
f (1) 0
Figure 2.28. Global reaction rate

f (u) as a function of reactant con-
centration u for various k2,f /k1,f val-
ues: curve a, 0.08; curve b, 0.06;
curve c, 0.05. Reaction system: paral-
lel nonisothermal reactions following
Langmuir–Hinshelwood kinetics, ac-
cording to equation (2.181); β1 = 0.3,
β2 = 0.3, γ1 = 20, γ2 = 35, σ1 = 8,
σ2 = 0.02, ε1 = 0, ε2 = 50, m1 = m2 =
2. (From Baratti et al., 1997.)
while satisfying the constraints (2.127) and (2.128). Equations (2.135)–(2.137) now
become
du
= γ s −n (2.185)
ds
dγ
= φ 2 µ(s) f (u)s n (2.186)
ds
dh
= µ(s)s n . (2.187)
ds
The corresponding BCs are
s = 0: γ =0 (2.188a)
s = 1: u=1 (2.188b)
s = 0: h=0 (2.188c)
1
s = 1: h= . (2.188d)
n+1
Using the overall effectiveness factor as the objective functional, from equation
(2.139) the Hamiltonian is given by
H = µ(s) + λ1 γ (s)s −n (2.189)
where the switching function is
= [(1 + λ2 )φ 2 f + λ3 ] s n . (2.190)
Similarly, the Lagrange multipliers λ1 , λ2 , and λ3 are defined as [cf. equations

(2.140)–(2.142)]
dλ1 ∂f
= −φ 2 (1 + λ2 ) µ(s)s n (2.191)
ds ∂u
dλ2
= −λ1 s −n (2.192)
ds
dλ3
=0 (2.193)
ds
with the BCs
s = 0: λ1 = 0 (2.194a)
s = 1: λ2 = 0. (2.194b)
Since the Hamiltonian is linear with respect to µ(s), the optimal catalyst loading
profile is the following step function (see also section 2.5.3):

α when > 0
µ= (2.195)
0 when < 0.
The number and location of steps depend on the specific form of the switching
function, whose evaluation requires complete solution of the problem. This has
Figure 2.29. Optimal active element distribution µ (continuous curve) and

reactant concentration u (dashed curve) within the catalyst pellet for var-
ious values of the upper bound α on the dimensionless active element
concentration. The values of the corresponding effectiveness factors are
also indicated. Reaction system: parallel nonisothermal reactions following
Langmuir–Hinshelwood kinetics, according to equation (2.181); β1 = 0.3,
β2 = 0.3, γ1 = 20, γ2 = 35, σ1 = 8, σ2 = 0.02, ε1 = 0, ε2 = 50, m1 = m2 = 2,
φ = 3, k2,f /k1,f = 0.06. (From Baratti et al., 1997.)
been carried out numerically, using the method described in section 2.5.4, and the
results are shown in Figure 2.29 for a few selected values of the upper bound α on
the dimensionless catalyst loading.
For low values of α, the optimal distribution is given by a single-step function.
As α increases, it becomes possible to concentrate more catalyst in a given support
area. In particular, for the cases shown in Figure 2.29.c and d, the catalyst has been
placed in two locations, which (as can be seen from the corresponding average
values of u) are near to those where f (u) exhibits its maxima. In these cases, the
optimal distribution is given by a two-step function. When α is further increased
(α > 7.4), the optimal distribution again takes the form of a single-step function,
since α is now large enough so that all the catalyst can be located in the vicinity
of the point where its utilization is maximized, i.e. where f (u) exhibits the larger
of its maxima. In the present case (see Figure 2.28), this corresponds to the lower
value of u, and hence to the inner of the two steps noted above. In the limit as
α → ∞, the optimal distribution approaches a Dirac-delta function.
In the example discussed above, the appearance of a two-step optimal catalyst

distribution is due to the fact that the overall reaction rate exhibits two local max-
ima with respect to reactant concentrations. This observation leads to the general
conclusion that, in practical applications, the optimal active catalyst distribution
can exhibit the multiple-step form only in cases where the involved reactions have
different kinetics. Since we are considering here the case where the catalyst is the
same, this implies that the two reactions should exhibit different mechanisms, i.e.
involving the breaking of different chemical bonds. Examples include isomeriza-
tion and hydrogenation of an unsaturated hydrocarbon, and parallel exothermic
reactions with different kinetics and activation energies (Baratti et al., 1997).
The practical implication of these results is that, particularly when dealing with
complex systems, as in most industrial applications, one should not assume a priori
that the optimal catalyst distribution has the form of a single step. It can in fact
well be a multiple-step distribution.
2.6 Experimental Studies

In addition to the considerable amount of theoretical research, several experimen-
tal studies have been conducted which demonstrate the improved performance of
nonuniformly distributed catalysts. In this context, a substantial volume of work
is related to automotive exhaust catalysts, and will be discussed in section 6.1. The
experimental research, in general, shows good qualitative, and in certain cases
also quantitative, agreement with the theory.
2.6.1 Oxidation Reactions

Kotter and Riekert (1983a,b) investigated the triangular reaction network of par-
tial oxidation of propene to acrolein on CuO/α-Al2 O3 . Three types of catalyst were
prepared, having approximately the same amount of CuO, deposited in uniform,
eggshell, and egg-yolk distributions. They found that the selectivity to acrolein was
higher for the eggshell catalyst and progressively lower in the uniform and egg-yolk
catalysts, in good agreement with theoretical predictions for the reaction network.
On Pt/γ -Al2 O3 catalysts prepared by coimpregnation techniques, Chemburkar
(1987) studied carbon monoxide oxidation, which in excess oxygen exhibits a max-
imum in the rate as a function of CO concentration. Steady-state multiplicity be-
havior was observed, and it was found that the catalyst performance was sensitive
to even small catalytic activity present in the portion of the support surrounding
the active layer (see Figure 2.30). In agreement with this study, Harold and Luss
(1987) demonstrated theoretically that multiplicity features for carbon monoxide
oxidation are altered significantly by small variations in the activity profile of an
eggshell-type catalyst.
Gavriilidis and Varma (1992) studied the ethylene epoxidation reaction net-
work on Ag/α-Al2 O3 step catalyst in a single-pellet reactor with external recycle.
As shown in Figure 2.31, the selectivity to ethylene oxide is higher when the active
material is located as a thin layer at the external surface of the pellet. These results,
Figure 2.30. Simulations using the activity profile illustrated

in the inset, for CO oxidation on Pt/γ -Al2 O3 , for various val-
ues of a/b. The circles denote experimental data. The curves
represent solutions of the diffusion–reaction model. (From
Chemburkar, 1987.)
which were obtained under excess oxygen are, consistent with the detrimental ef-
fect of temperature gradients on selectivity (Morbidelli et al., 1984a; Pavlou and
Vayenas, 1990a), owing to the higher activation energy of ethylene combustion
than of ethylene epoxidation. Under ethylene-rich conditions, in addition to tem-
perature gradients, oxygen concentration gradients affect the selectivity negatively
(Yeung et al., 1998). Reactant concentration gradients are also detrimental for
Figure 2.31. Selectivity to ethylene oxide during ethylene

epoxidation on Ag/α-Al2 O3 , as a function of active layer
location, for various bulk fluid temperatures; inlet ethy-
lene concentration 7.2%. (From Gavriilidis and Varma,
1992.)
Figure 2.32. Ethylene oxide selec-

tivity during ethylene epoxidation
on Ag/α-Al2 O3 , as a function of
DCE concentration at a fixed 10%
ethylene conversion for various dis-
tributions of catalyst: (a) no ethane,
and (b) 0.5% ethane. (From Yeung
et al., 1998.)
conversion. As a result, a surface Dirac-type catalyst pellet exhibits high ethylene

conversion as well as high selectivity and yield towards ethylene oxide.
When dichloroethane (DCE) is added to the feed as a promoter, the situation
changes (Yeung et al., 1998). In general, subsurface catalyst distributions have the
best selectivity and yield, as shown in Figure 2.32.a for 10% ethylene conversion.
DCE inhibits both the epoxidation and combustion reactions. In particular, low
concentrations of DCE improve selectivity, while higher concentrations are detri-
mental due to formation of bulk silver chloride. A surface Dirac-type catalyst has
better performance at low DCE levels (≤0.2 ppm), while enhanced selectivity is
attained at higher DCE levels (0.7–1.5 ppm) for subsurface Dirac catalysts. The
observed effects are due to transport resistances, which decrease the concentra-
tion of DCE along the depth of the pellet. An active layer placed at subsurface
locations is exposed to a lower concentration of DCE than is a surface catalyst.
This gives rise to the desired inhibition effect that enhances selectivity.
Experiments were also performed with ethane present in the feed. Ethane
reacts with adsorbed chlorine, thus preventing its accumulation at the catalyst
surface. Similar behavior was obtained (see Figure 2.32.b), but the maximum
displayed by the subsurface Dirac catalyst was broader, allowing a larger window
of operation and greater tolerance for fluctuations in the feed DCE concentration.
Placing the active layer at a subsurface location was shown also to significantly
reduce the amount of trace contaminants present in the feed gas that reach the
silver catalyst.
2.6.2 Hydrogenation Reactions

Masi et al. (1988) studied ethylene hydrogenation in an external recycle reactor, on
Pt/γ -Al2 O3 catalysts prepared by coimpregnation techniques. Steady-state multi-
plicity behavior was observed. For a range of ethylene concentration values, larger
reaction rates were obtained by subsurface than by surface distribution. This fea-
ture was attributed to the fact that the hydrogenation reaction rate exhibits a
maximum as a function of ethylene concentration.
Wu et al. (1988) investigated the hydrogenation of ethylene on Pd/γ -Al2 O3 ,
which follows first-order kinetics, and the methanation of CO on Ni/γ -Al2 O3 ,
which follows bimolecular Langmuir–Hinshelwood kinetics, in a single-pellet in-
ternal recycle reactor. The pellets were prepared by pressing the active catalyst
between two alumina layers, so that step distributions at any desired location could
be produced. It was found for both reactions that there exists an optimal location
for the active layer within the pellet (see Figure 2.33), which moves towards the ex-
ternal surface as the Thiele modulus increases, according to theoretical predictions.
Au et al. (1995) studied benzene hydrogenation on Ni/kieselguhr catalyst in a
single-pellet diffusion reactor. The kinetics of this reaction were correlated by an
Figure 2.33. (a) Plot of experimental ef-

fectiveness factor η vs active-layer lo-
cation s̄, for ethylene hydrogenation
on Pd/Al2 O3 : temperature 298 K, hy-
drogen concentration 5.89%, = 0.05.
(b) Experimental and calculated (dashed
line) dependences of effectiveness fac-
tor η on the active layer location s̄ for
CO methanation on Ni/Al2 O3 : tempera-
ture 533 K, CO concentration 5.0 × 10−6
mol/ml, H2 /CO = 7.85. (From Wu et al.,
1988.)
Eley–Rideal rate equation, which is of positive order with respect to both reac-
tants. Five different catalyst distributions ranging from preferential shell loading
to preferential core loading were examined under isothermal conditions. The ef-
fectiveness factor was found to be higher for pellets with the catalyst concentrated
towards the external surface, which is consistent with the Eley–Rideal kinetics.
At small Thiele modulus (i.e. kinetic control), all pellets showed the same perfor-
mance as expected, but the shell-type pellets showed notable improvements for
larger Thiele modulus values.
In complex hydrogenation reactions, selectivity is the performance index of
interest. In these cases, partially hydrogenated products are typically desired.
Komiyama et al. (1997) studied liquid-phase batch hydrogenation of 1,3-butadiene
over Pt/γ -Al2 O3 catalysts, which is a triangular reaction network
Butadiene → Butene
(2.196)
Butane
where butene hydrogenation is inhibited in the presence of butadiene. Uniform,

eggshell, and egg-white catalyst pellets were used, and the highest selectivities
to butenes were obtained with the eggshell distribution. The residence time of
butenes inside the catalyst pellet was larger for uniform and egg-white pellets,
thus providing a greater chance to react further to butane. For the uniform cata-
lyst, an interesting feature observed was that the butane fraction decreased with
time during the initial stages of the reaction. This was attributed to the fact that
the contribution of butene hydrogenation to butane production is larger in the
beginning, due to the absence of butadiene in the pellet interior.
A similar situation exists in the case of acetylene hydrogenation on Pd/Al2 O3 .
Even if ethylene is present at high concentration, 100% selectivity to ethylene can
be obtained by the optimal catalyst profile (eggshell), because traces of acetylene
can poison ethylene hydrogenation completely (Mars and Gorgels, 1964). There-
fore, a zone in the catalyst layer which is free of acetylene is detrimental, because
ethylene can be hydrogenated in this zone. For this reason, the best catalyst dis-
tribution is a sharp eggshell that minimizes acetylene concentration gradients.
Uemura and Hatate (1989) studied the same reaction on Ni/Al2 O3 pellets with
thin eggshell, thick eggshell, uniform, and egg-yolk profiles. They obtained the
highest selectivity using the thin eggshell catalyst, as shown in Figure 2.34, in
agreement with theoretical calculations. However, even in the absence of ethylene
in the feed, the highest selectivity was about 80%, obtained at low conversion
using the thin eggshell catalyst. This was attributed to the existence of a direct
hydrogenation route from acetylene to ethane.
Berenblyum et al. (1986) investigated the selective hydrogenation of C5 –C8
dienes to alkenes on eggshell Pd catalysts with different catalyst loadings (0.05–
0.5 wt%) and shell thicknesses (0.1–1.5 mm). They observed that the highest diene
conversion was obtained at intermediate values of catalyst loading and thickness
(0.25% Pd and 0.2-mm catalyst thickness), while pellets with active layer thick-
ness less than 0.2 mm and loadings 0.15–0.25% Pd offered the best selectivities.
Figure 2.34. Ethylene selectivity as

a function of acetylene conversion
during acetylene hydrogenation on
various nonuniform Ni/Al2 O3 pel-
lets. T = 413 K, PA = 5.1 kPa, PH =
4.1 kPa. (From Uemura and Hatate,
1989.)
This behavior was attributed to an interaction of diffusion–reaction phenomena,

loading dispersion dependence, and structure sensitivity of the reactions.
2.6.3 Fischer–Tropsch Synthesis

Eggshell Co/SiO2 catalysts were shown to be more active and selective than uni-
form pellets for C+ 5 production during Fischer–Tropsch synthesis (Iglesia et al.,
1993; Iglesia et al., 1995; Iglesia, 1997), because they decrease diffusional restric-
tions that normally lead to low reaction rates and selectivities. However, a certain
level of diffusional resistance was found to be beneficial to selectivity. Pellets with
intermediate thickness of the catalyst layer provided maximum selectivity, be-
cause they avoid large intraphase CO concentration gradients while restricting
the diffusive removal of reactive olefin products, which can continue to grow to
higher molecular weight hydrocarbons. The higher C+ 5 selectivity of eggshell cat-
alysts than of uniform ones was also reported previously in the patent literature
(van Erp et al., 1986; Post and Sie, 1986).
3
Optimization of the Catalyst
Distribution in a Reactor
I
n single-pellet studies, it is assumed that no concentration or temperature gradi-
ents are present in the fluid phase surrounding the pellet. A reactor with external
or internal recycle is one of the experimental realizations of the single-pellet
concept. However, in a fixed-bed reactor, the fluid-phase composition and temper-
ature vary with position. For this reason, the optimization problem becomes more
complex. Thus, it is not surprising that relatively few reactor studies have appeared
in the literature as compared with those for single pellets. In this chapter, we first
discuss theoretical studies of single and multiple reactions under isothermal and
nonisothermal conditions, and then present experimental work which supports
the theoretical developments.
3.1 A Single Reaction

One of the earliest works in this area considered CO oxidation in excess oxygen
over platinum catalyst, in monolith reactors for automobile converters (Becker
and Wei, 1977a). Recall that this reaction exhibits a maximum in the rate as a
function of CO concentration. The CO conversion to CO2 as a function of the
Thiele modulus (or equivalently, the catalyst temperature) for a fixed inlet CO
concentration is shown in Figure 3.1 for four monoliths with different catalyst
distributions (cf. Figure 2.1). The conversion of CO below 650◦ F was significantly
higher in the inner catalyst, while above 650◦ F the middle catalyst performed
better and attained 100% conversion. The uniform and outer catalysts also showed
100% conversion at progressively higher temperatures.
The optimization of the isothermal fixed-bed reactor, where a bimolecular
Langmuir–Hinshelwood reaction occurs, was performed analytically by Morbidelli
et al. (1986a,b). In this case, the mass balance for a single pellet [equation
(2.2)] must be coupled with the mass balance for the fluid phase, which for a
69
70 Optimization of the Catalyst Distribution in a Reactor
Figure 3.1. Carbon monoxide conversion to CO2 over a mono-

lith reactor with various nonuniform distributions of platinum
catalyst. (From Becker and Wei, 1977a.)
heterogeneous plug-flow reactor is

dCf
v = −(1 − ε)r̄ (Cf ) (3.1)
dz
z = 0: C f = Cf0 (3.2)
where r̄ (Cf ) represents the volume-averaged reaction rate in the catalyst phase
and is expressed by

1
r̄ (Cf ) = r (C) a(x) dVp . (3.3)
Vp Vp
The dimensionless form of equation (3.1) is

1
dg
= −Da (n + 1) f (u) a(s, y)s n ds (3.4)
dy 0
y = 0: g=1 (3.5)
where the following dimensionless quantities have been introduced:
g = Cf /Cf0 , y = z/L, Da = L(1 − ε)r (Cf0 )/vCf0
r (C) (1 + σ )2 u (3.6)
u = C/Cf0 , σ = KCf0 , f (u) = = .
0
r (Cf ) (1 + σ u)2
The fluid-phase balance has to be solved in conjunction with the solid-phase bal-
ance [equations (2.2) and (2.5)], which using the above dimensionless variables
reduces to

1 d n du
s = φ 2 a(s) f (u) (3.7)
s n ds ds
with BCs
du
s = 0: =0 (3.8a)
ds
s = 1: u=g (3.8b)
where the second BC allows for the fact that the fluid-phase composition changes
along the reactor length.
The objective of reactor optimization is to maximize the conversion at the
reactor outlet,
X = 1 − g(1). (3.9)
The conversion can be evaluated by integrating the external fluid-phase mass

balance to give
1
X = Da e(y) dy (3.10)
0
where
1
e(y) = (n + 1) f (u)a(s, y)s n ds (3.11)
0
represents the contribution to reactant conversion at location y along the reactor

axis. The optimization variable is the activity distribution a(s, y), which now de-
pends also on the position along the reactor axis y, because each particle along
the bed may have a different distribution. Since e(y) is always positive, maximum
conversion is obtained by locally maximizing e(y) at each y ∈ [0, 1]. Thus, the
problem is reduced to optimizing the single-pellet performance at each position.
The difference between the reactor and the single-pellet case is in the BC for u,
which now varies with position along the reactor, i.e., u(1, y) = g(y) ≤ 1. Using
the same arguments as in the pellet case, it can be shown that the optimal distri-
bution is a Dirac-delta function centered at a location s̄(y) within the pellet such
that u(s̄) = um , that is, where the concentration is equal to the value which maxi-
mizes the reaction rate. The method of finding the optimal catalyst location along
the reactor by single-pellet optimization may be referred to as the local optimal
catalyst distribution (OCD) approach.
Solving the pellet mass balance with the Dirac catalyst distribution
δ[s − s̄(y)]
a(s, y) = (3.12)
(n + 1)s̄ n (y)
Table 3.1. Location of optimal catalyst distribution along a

reactor and corresponding local consumption rate for
isothermal, bimolecular Langmuir–Hinshelwood kinetics, without
external transport resistances.
σ < 1 or
σ > 1 and g < 1/σ σ > 1 and g > 1/σ
s̄opt (y) 1 1 − a n=0
exp(−) n=1
1
n=2
1+
(1 + σ )2
eopt (y) f (g) f (um ) =
4σ
a For > 1, s̄opt = 0 and eopt (y) = f (ū), where ū is the solution of 1 − ū −
φ 2 f (ū) = 0.
the following expressions for the optimal location are derived:
s̄opt (y) = 1 − for n=0 (3.13a)

s̄opt (y) = exp(−) for n=1 (3.13b)
1
s̄opt (y) = for n=2 (3.13c)
1+
where
4(n + 1)(σ g − 1)
= (3.14)
φ02
so that, since g = g(y), s̄opt is a function of reactor position y. The resulting value of
the local consumption rate, e(y), is obtained by substituting the optimal Dirac dis-
tribution in equation (3.11). Note that these solutions do not hold for the cases σ <
1, or σ > 1 and g < um = 1/σ . In these cases the reaction rate f (u) is monoton-
ically increasing in the interval u ∈ [0, g] and attains its maximum value at u = g.
Therefore, the optimal catalyst location is at the pellet external surface, i.e. s̄opt = 1.
Note also that for the infinite slab (n = 0), the value of s̄opt can become negative.
In this case, as discussed in section 2.1.1, the optimum catalyst location is s̄opt = 0.
The expressions for the optimal location along the reactor and the correspond-
ing outlet reactant conversion are summarized in Table 3.1. They indicate that,
starting from the reactor inlet, until the fluid reactant concentration reaches 1/σ ,
the optimal pellets have subsurface Dirac-delta distribution. The catalyst location
moves progressively towards the pellet external surface, while each pellet exhibits
constant local reaction rate, eopt .
When the fluid reactant concentration becomes lower than 1/σ , then beyond
this reactor position, the pellets have surface Dirac distribution and the local
reaction rate changes with fluid reactant concentration. The fluid concentration
g and the optimal location s̄opt are illustrated in Figure 3.2 for an optimal reactor
Figure 3.2. Rigorous optimal solution: profiles of the optimal active catalyst
location s̄ opt and the bulk fluid concentration g along the reactor axis y for
a bimolecular Langmuir–Hinshelwood reaction. (From Morbidelli et al.,
1986b.)
where g(y) > 1/σ . Integrating equation (3.4) yields the fluid concentration profile
g(y) = 1 − Da eopt y. (3.15)
Note that in this case, only the optimal catalyst location s̄opt depends on the Thiele
modulus, while the g profile remains unchanged, since diffusional resistances affect
only the location inside the pellet where um is realized.
The obtained optimal solution requires the catalyst location to be different in
each pellet, depending on its location along the reactor. This is difficult to realize
in practice, but provides the theoretical limit of reactor performance, and hence
may be referred to as the absolute optimum solution. The more practical situa-
tion where the catalyst location is the same for all pellets has also been examined
(Lee et al., 1987a). Its performance was only slightly below the absolute optimum
solution for low conversion, but became progressively worse for higher conver-
sions. As an alternative configuration, the fixed-bed is divided into multiple zones,
where in each zone the catalyst location is the same but different from the others.
The optimization variables are now the catalyst location within pellets in each
zone, as well as the length of the corresponding zone. As shown in Figure 3.3,
for any given Damköhler number Da, and for increasing number of zones, d, the
optimized reactor configuration evolves towards the absolute optimum, achieved
for an infinite number of zones, d = ∞. By the use of only two or three zones, the
theoretical optimal reactor performance is reasonably closely approached. The
optimal catalyst locations and zone lengths for reactors having one, two, three, or
an infinite number of zones are illustrated in Figure 3.4.
In reactor optimization studies, care should be exercised for the possible oc-
currence of multiple steady states. For the reaction system considered above,
Figure 3.3. Comparison of reactor

outlet conversion X for various num-
ber of zones in the reactor, d, as a
function of Damköhler number Da.
Bimolecular Langmuir–Hinshelwood
reaction, σ = 20. (From Lee et al.,
1987a.)
reactor steady-state multiplicity was observed (three steady states, of which two
were stable) and attributed to the multiplicity behavior of the first layer of pellets
packed along the axial direction (Morbidelli et al., 1986a). Criteria were developed
to predict the existence of multiple steady states and ignition conditions. In the
case of a single-zone reactor with axial dispersion, the number of possible steady
states increases to nine, among which only three are stable (Lee et al., 1987b).
Two sources of multiplicity are present in this case: the heterogeneous nature of
the reactor and the presence of axial dispersion in the fluid phase.
Homogeneous–heterogeneous reactions
In the cases discussed up to now, only catalytic reactions have been considered.
However, there are certain situations where both heterogeneous and homoge-
neous reactions take place, with some species participating in both types of re-
actions. One such case is catalytic combustion, where heterogeneous reactions
can enhance homogeneous ones because the catalyst promotes the production of
free radicals, which can increase the rate of homogeneous reactions (Pfefferle and
Pfefferle, 1987). The simple case of two first-order reactions, one heterogeneous
and the other homogeneous, was studied by Melis et al. (1997). The analysis was
carried out in the absence or presence of interactions between the two reaction
Figure 3.4. Optimal catalyst location s̄ opt along the reactor axis y for various num-
bers of zones in the reactor, d. Bimolecular Langmuir–Hinshelwood reaction, σ =
5, Da = 0.3, φ 2 = 0.7, n = 1. (From Lee et al., 1987a.)
types, under isothermal conditions, using reactor conversion as the optimization

parameter. Enhancement of homogeneous reactions was assumed to take place
only in the noncatalytic portion of the reactor and was allowed for in a simple way,
by a discrete increase of the apparent rate constant of the homogeneous reaction
through an enhancement parameter. A zone reactor was considered, with each
zone either containing catalyst or not. In catalyst-containing zones, the catalyst
distribution was uniform both radially and axially.
In the absence of enhancement of homogeneous reactions, the optimal reactor
was found to be a one-zone reactor containing catalyst, except for the case of
negligible mass transfer resistance, where conversion was independent of cata-
lyst distribution. When the presence of the catalyst enhanced the homogeneous
reactions, it was found that a two-zone reactor was optimal, with the first zone
catalytic and the second noncatalytic. When the mass transfer resistance was large
and for sufficiently high values of the enhancement factor, the optimal policy
was to support the homogeneous reaction to the greatest extent. This occurred
with a Dirac-type first (catalytic) zone, as shown in Figure 3.5.a,b. As the mass
transfer resistance decreased, as the intrinsic catalytic activity increased, or as the
enhancement of homogeneous reaction became weaker, the length of the first
catalytic zone increased at the expense of the noncatalytic one, which eventually
disappeared, as shown in Figure 3.5.c,d, because the contribution of the heteroge-
neous reactions became more prominent. Even though the above study is limited
to a simple case, it demonstrates that when homogeneous reactions are present
the optimal catalyst distribution can be greatly affected.

For a single reaction taking place in an adiabatic reactor, the steady-state fluid-
phase mass and energy balances for a heterogeneous plug-flow reactor are (Lee
and Varma, 1987)
dCf
v = −(1 − ε)r̄ (Cf , Tf ) (3.16)
dz
dTf
ρcp v = (1 − ε)(−H)r̄ (Cf , Tf ) (3.17)
dz
z = 0: Cf = Cf0 , Tf = Tf 0 . (3.18)
The volume-averaged reactivity of the solid catalyst phase, r̄ (Cf , Tf ), is given by

equation (3.3), where now the reaction rate r is also a function of temperature.
The dimensionless mass and energy balances are
1
dg
= −Da (n + 1) f (u, θ) a(s, y)s n ds (3.19)
dy 0
1
dτ
= Da (n + 1)β̄ f (u, θ) a(s, y)s n ds (3.20)
dy 0
Figure 3.5. Reactor exit conversion X as a function of dimensionless length of the first
(catalytic) zone when the homogeneous reaction rate in the second (noncatalytic) zone is
enhanced through the catalytic reactions by the enhancement parameter shown. Da = 1.
Dimensionless mass transfer coefficient, dimensionless heterogeneous reaction rate con-
stant: (a) 1, 1; (b) 1, 5; (c) 5, 1; (d) 5, 5. (From Melis et al., 1997.)
y = 0: g = 1, τ =1 (3.21)
where the following dimensionless quantities have been used:

Tf (−H)Cf0 L(1 − ε)r Cf0 , Tf 0
τ = 0, β̄ = , Da =
Tf ρcp Tf 0 vCf0
(3.22)
T r (C, T)
θ = 0, f (u, θ ) = 0 0 .
Tf r Cf , Tf
The above fluid-phase balances need to be solved in conjunction with the solid-
phase balances (2.43)–(2.45), which using the above dimensionless variables take
the form

1 d du
sn = φ 2 a(s, y) f (u, θ) (3.23)
s n ds ds

1 d n dθ
s = −βφ 2 a(s, y) f (u, θ ) (3.24)
s n ds ds
du dθ
s = 0: = 0, =0 (3.25a)
ds ds
s = 1: u = g, θ = τ. (3.25b)
When considering conversion optimization, it is evident from equation (3.19)

that optimality is achieved by maximizing the reaction rate at all locations along
the reactor. This means that, as for the isothermal reactor, optimum performance
is achieved with Dirac-delta catalyst distributions centered at different locations
depending on pellet position along the reactor.
The steady-state multiplicity behavior of such a reactor can be quite complex.
Using singularity theory, Lee and Varma (1987) studied the multiplicity charac-
teristics of an adiabatic plug-flow reactor with a single zone of Dirac-type cata-
lysts for bimolecular Langmuir–Hinshelwood and mth-order reactions. Due to the
strongly nonlinear solid–fluid interactions, a large variety of bifurcation diagrams
was found in this case.
3.2 Multiple Reactions

In the previous section, reactor performance was optimized for single reactions
using the local OCD approach. This implies that the activity distribution in each
particle along the reactor is optimized independently, so as to promote locally the
best attainable performance. In the case of multiple reactions this approach may
not be adequate.
This has been demonstrated for an isothermal parallel reacting system (Wu,
1994)
A1 → A2
(3.26)
A1 → A 3
where A2 is the desired product. The kinetic expressions considered for the two
reactions are
k1 C1m1
r1 = (3.27a)
(1 + K1 C1 )b1
k2 C1m2
r2 = . (3.27b)
(1 + K1 C1 )b2
The dimensionless reactor mass balances are

2 1
dgi
= νij Da j (n + 1) f j (u) a(s, y)s n ds, i = 1, . . . , 3 (3.28)
dy j=1 0
which for a Dirac catalyst distribution become
dgi 2
= νij Da j f j (ū). (3.29)
dy j=1
Since the reaction rates (3.27) depend only on the concentration of a single com-
ponent, the vector u is in reality simply u1 .
The reactor selectivity and yield are defined by
g2 (1)
S= (3.30)
1 − g1 (1)
Y = g2 (1) = X S (3.31)
where X is the reactor conversion. According to the local OCD approach, the
optimal value of ū1 is determined as the one which locally maximizes the desired
reactor performance. In the case where such a value of ū1 is larger than the cor-
responding concentration value in the bulk, g1 , the optimal performance is not
obtainable and we have the suboptimal situation where ū1 = g1 and s̄ = 1. It is
in this event that the global OCD approach can provide better performance. In
this case the ū1 value is obtained numerically by a one-parameter optimization
procedure where now the objective function is not the local but rather the global
reactor performance. In Figure 3.6, with reference to the reacting system above, it
is shown that the global OCD approach can result in higher reactor yield than the
local approach. A difference in the performance index obtained by the two meth-
ods exists only when the reactor has two sections: one with s̄opt < 1 (0 < y < ŷ)
and the second with s̄opt = 1( ŷ < y < 1). The difference increases when the length
of the second section (1 − ŷ) increases. Similarly, the global OCD approach can
also give higher reactor selectivity than the local OCD approach (Wu, 1994).
An alternative design approach, which utilizes a one-zone reactor of Dirac cat-
alysts but still results in the same optimum performance obtained by the global
OCD approach, has been proposed (Sheintuch et al., 1986; Wu, 1995). This was
accomplished by optimizing the feed distribution along the reactor. In particu-
lar, Wu (1995) considered the case of two isothermal parallel reactions following
Langmuir–Hinshelwood kinetics (3.27), with yield as the performance index. Uti-
lizing the maximum principle, the optimal feed distribution was found analytically
to be a uniform distribution of the feed stream along a reactor portion starting
from its entrance, so the optimization problem was reduced to determining the
length of this portion and the optimal catalyst location. For low Damköhler val-
ues, the feed distribution portion was the whole reactor length and the associated
optimal catalyst location was a subsurface position. For large Damköhler values,
the feed distribution portion was shorter than the reactor length and the optimal
catalyst location was the pellet external surface.
Figure 3.6. Maximum reactor

outlet yield Y and reactor portion
with subsurface optimal cata-
lyst location, ŷ, as a function of
Damköhler number Da. Reaction
system given by equations (3.27),
m1 = 1, b1 = 2, m2 = 1, b2 = 1, σ = 20,
n = 2, φ12 = 0.2, κ = 1. Global OCD
approach (solid curves), local OCD
approach (dashed curves). (From
Wu, 1994.)
A comparison of the optimal feed distribution (OFD) and the global OCD
approaches is shown in Figure 3.7. Both approaches lead to the same outlet yield,
and the length of the feed distribution for the OFD approach is equal to the
length of the reactor portion with subsurface optimal catalyst location for the
OCD approach. These similarities are not coincidental and are directly related to
the fact that the ū1 profile along the reactor is the same. For the OCD approach
this profile is achieved by adjusting the catalyst location in pellets at each point
along the reactor; for the OFD approach it is achieved by properly distributing
the feed. Therefore the two approaches give the same reactor performance and
differ only in the way they achieve it.

One- and two-zone reactor optimization has been reported for the case of ethylene
epoxidation on silver catalyst taking place in a nonisothermal, nonadiabatic fixed-
bed reactor (Baratti et al., 1994). The reaction scheme considered is
1
C2 H4 + 12 O2 −→ C2 H4 O
(3.32)
2
C2 H4 + 3O2 −→ 2CO2 + 2H2 O
which follows the reaction kinetics (Klugherz and Harriott, 1971)
k1 C1 C32
r1 = √ (3.33a)
(0.0106 + 2144C1 + 805C3 )(1 + 1271 C3 )
k2 C1 C32
r2 = √ (3.33b)
(0.008 + 4166C1 + 1578C3 )(1 + 718 C3 )
where C1 and C3 are the concentrations (mol/cm3 ) of ethylene and oxygen, re-
spectively. This is a parallel reaction scheme similar to (3.26). Yield and selectivity
Figure 3.7. (a) Optimal dimen-

sionless feed distribution along
the reactor, wopt (y), corresponding
dimensionless fluid-phase concentra-
tions g1 and g2 , and dimensionless
concentration at the catalyst loca-
tion, ū1 . The catalyst location is
at the external surface of pellets.
Outlet yield maximized using the
OFD approach. (b) Optimal catalyst
location s̄ opt , corresponding dimen-
sionless fluid-phase concentrations
g1 and g2 , and dimensionless con-
centration at the catalyst location,
ū1 . Outlet yield maximized using
the global OCD approach. Reac-
tion system given by equations (3.27),
m1 = 1, b1 = 2, m2 = 1, b2 = 1, σ = 20,
n = 2, φ12 = 0.2, κ = 0.5, Da = 0.25.
(From Wu, 1995.)
used as the performance indexes are given by equations (3.30) and (3.31), where
g1 and g2 are the dimensionless bulk fluid concentrations of ethylene and ethy-
lene oxide, respectively. The fluid-phase mass and energy balances for a one-zone
reactor are
2 1
dgi
= νij Da j (n + 1) f j (u, θ ) a(s)s n ds (3.34)
dy j=1 0
2 1
dτ
= β̄ j Da j (n + 1) f j (u, θ ) a(s)s n ds − St(τ − τc ) (3.35)
dy j=1 0
which for a Dirac catalyst distribution become
dgi 2
= νij Da j f j (ū, θ̄ ) (3.36)
dy j=1
dτ 2
= β̄ j Da j f j (ū, θ̄) − St (τ − τc ) (3.37)
dy j=1
where St is the Stanton number. The solid-phase mass and energy balances (2.94)–
(2.96) were reduced to algebraic equations using standard techniques. Optimiza-
tion was carried out numerically, through one- or multiparameter estimation meth-
ods for the one- and two-zone reactors, respectively.
For small values of the Thiele modulus, the optimal catalyst location for yield
maximization was mostly in the pellet interior. In this case, no discernible dif-
ference in the performance of the optimal, surface, and uniform distributions
was observed, because kinetic control prevailed. Increasing catalyst activity led to
ignition behavior for both surface and uniform catalysts. For the optimal Dirac dis-
tribution, though, intraparticle gradients present for subsurface catalysts helped
to avoid reactor runaway. Thus, when using Dirac catalysts it is possible to operate
the reactor with higher loading (activity) catalysts, while avoiding reactor runaway.
Figure 3.8. Dimensionless fluid-phase

concentrations for ethylene (g1 ) and
oxygen (g2 ) and temperature rise T
along the reactor for φ = 0.5 for (a)
uniform catalyst, (b) surface catalyst,
(c) optimal catalyst for yield maximiza-
tion in ethylene epoxidation (s̄opt = 0.8).
(From Baratti et al., 1994.)
This has a beneficial effect on reactor performance, since while ignition results in
low yields for the uniform and surface distributions, the yield obtained with the
optimal Dirac catalyst continues to increase with catalyst activity. This behavior is
clearly demonstrated in Figure 3.8, where dimensionless bulk-fluid concentrations
and temperature rise (above the inlet reactor temperature) along the reactor are
shown for the three distributions, using parameter values representative of typical
reaction conditions. The surface pellet ignites earlier than the uniform, because
there are no intraphase concentration gradients to slow down reaction rates. The
runaway situation is not observed for the optimal Dirac distribution. In this case,
there is a hot spot close to the reactor entrance, but eventually the temperature
stabilizes.
For selectivity maximization only two positions were found to be optimal, de-
pending primarily on the value of Thiele modulus: surface and center. This behav-
ior is directly related to the complexity of the intrinsic reaction kinetics. However,
subsurface locations were associated with negligible ethylene conversion, so that
the surface distribution is the only one of interest from the practical point of
view.
In Figure 3.9, a comparison between one- and two-zone reactors is shown for
yield maximization. The optimal catalyst location in the second zone, s̄2,opt , is larger
than in the first, s̄1,opt , in order to compensate for the decreased bulk reactant
concentration. The optimal location s̄opt for the one-zone reactor falls between
s̄1,opt and s̄2,opt , since it is in a sense their average. As the catalyst activity increases,
the optimal location for the second zone reaches the pellet surface. It does not
Figure 3.9. Comparison between one-

and two-zone reactors for yield max-
imization in ethylene epoxidation. (a)
Yield Y as a function of Thiele modulus
φ. (b) Optimal catalyst locations for one-
zone reactor (s̄opt ) and for two-zone re-
actor (s̄1,opt, s̄2,opt ) and dimensionless op-
timal length of the first zone, y1,opt , as
a function of Thiele modulus φ. (From
Baratti et al., 1994.)
move to subsurface locations as s̄opt and s̄1,opt do, because in the second zone
reactant concentrations are low, thus preventing runaway. The improvement by
using a two-zone reactor is insignificant at low values of Thiele modulus, but at
larger values yield improvements of about 15% over the single-zone reactor can
be realized.
The fact that optimization of catalyst distribution leads to avoidance of reactor
runaway in ethylene epoxidation is attractive. Since the desired reaction has lower
activation energy than the undesired reaction, excessive temperatures result in low
selectivity and yield. Hence, in order to maximize yield and selectivity, reactor run-
away must be avoided. It is worth noting that axial temperature gradients can also
be suppressed by using uniform pellets and diluting them appropriately with inert
pellets or by using uniform pellets but with catalyst loading depending on their
position along the reactor. In this way, a nonuniform activity profile is established
along the reactor length. Both approaches have been shown to prevent reactor
runaway and can also lead to improved selectivity (Caldwell and Calderbank,
1969; Narsimhan, 1976; Sadhukhan and Petersen, 1976; Pirkle and Wachs, 1987).

A few experimental studies describing the effects of nonuniform catalyst distribu-
tions in fixed-bed reactors have been reported in the literature. They are primarily
concerned with oxidation reactions and are described below. Additional related
studies in the context of automotive exhaust catalysis are reported in section 6.1.
3.3.1 Propane and CO Oxidation

Propane oxidation was investigated on Pt/γ -Al2 O3 catalyst by Kunimori et al.
(1982) in a fixed-bed reactor under isothermal conditions. They found that the
observed activity was as follows: eggshell > uniform > egg-white, which is expected
for first-order kinetics. However, in poisoning experiments, the egg-white catalyst
was more resistant, which is consistent with a pore-mouth poisoning mechanism.
These investigators also studied CO oxidation on the same catalyst. As noted
earlier, this reaction exhibits bimolecular Langmuir–Hinshelwood kinetics, and
egg-white catalysts can be more active. Indeed, the activity was in the order egg-
white > uniform > eggshell. Examining the steady-state multiplicity behavior of
the above system, it was found that the catalyst distribution alters the hysteresis
behavior (Kunimori et al., 1986). The diffusional resistance introduced by the
outer inert layer of the egg-white catalyst increases the width of the conversion-
vs-temperature hysteresis loop. It is interesting to note that no difference in activity
was observed between the powdered catalysts obtained by pulverizing the egg-
white and eggshell catalyst pellets, which indicates that the preparation procedure
did not alter the intrinsic catalyst activity.
Lee and Varma (1988) studied experimentally the behavior of an isothermal
fixed-bed reactor for CO oxidation. Four different Pt/γ -Al2 O3 catalysts with step
activity distribution were considered. The steps were very thin (not larger than 8%
Figure 3.10. CO oxidation on Pt/γ -Al2 O3 catalyst at

115◦ C and 152-ml/min flow rate. Comparison of one-
zone reactors: (a) s̄ = 0.27 and s̄ = 1.0, (b) s̄ = 0.65 and
s̄ = 0.50. In (a) results for the two-zone reactor (s̄1 =
0.27, s̄2 = 1.0) are also shown. (From Lee and Varma,
1988.)
of the pellet radius) and located approximately at s̄ = 0.27, 0.50, 0.65, 1. Two steady
states were observed in all single-zone reactors. As shown in Figure 3.10, among the
four types of catalysts studied, the reactor packed with the catalyst where the active
layer was positioned at s̄ = 0.65 had the widest range of high-conversion regime.
This range was further increased by using a dual-zone reactor, with s̄ = 0.27 and
s̄ = 1 in the first and second zones, respectively. This improvement was expected
according to the discussion in section 3.1.1. A direct experimental test of the
effect of nonuniform catalyst distribution was demonstrated in a dual-zone reactor.
Using the same catalyst pellets, the reactor performance (including the number of
steady states) was dramatically altered when the packing order was reversed, i.e.,
first s̄ = 0.27 and second s̄ = 1 in the two zones, or vice versa, as shown in Figure
3.11. All of the above observations were in excellent qualitative agreement with
theoretical predictions.
Figure 3.11. CO oxidation on Pt/γ -Al2 O3 catalyst at 100◦ C

and 152-ml/min flow rate. Comparison of two-zone reactors:
(a) s̄1 = 0.27 and s̄2 = 1.0, (b) s̄1 = 1.0 and s̄2 = 0.27. (From
Lee and Varma, 1988.)
3.3.2 Catalytic Incineration of Volatile Organic Compounds

Incineration of Volatile Organic Compounds was investigated by Frost et al. (1990)
in a nonisothermal fixed-bed reactor with axial mass and heat dispersion and
both inter- and intraparticle mass and heat transfer resistances, loaded with Pt/γ -
Al2 O3 pellets where the platinum was distributed as a thin layer slightly below
the external surface. The reaction was assumed to follow Langmuir–Hinshelwood
kinetics. The authors developed a model, which was fitted to experimental data on
conversion vs inlet temperature by optimizing the values of the kinetic parameters
and the effective diffusivity of reactants in the pellet. The model was used to size
commercial-scale catalytic incinerators.
4
Studies Involving Catalyst Deactivation
A
n important area where the techniques of catalyst design described above
can be used to improve process performance involves reacting systems whose
activity changes with time. These include catalysts undergoing poisoning as
well as processes where small amounts of undesired components are removed
from a stream (e.g. hydrodemetallation). In such catalytic systems a new variable
is introduced: the catalyst lifetime. It should be noted that only the initial catalyst
distribution can be controlled, while the performance index is usually integrated
over the catalyst lifetime. Next, we discuss catalyst design in the presence of two
deactivation mechanisms: nonselective and selective poisoning.
4.1 Nonselective Poisoning

Becker and Wei (1977b) studied the case of a first-order reaction with nonselective
poisoning. They assumed that the poison adsorbs on the entire catalyst surface with
no preference for the active element or the support (nonselective poisoning), and
without ever reaching saturation. The diffusion–reaction process for the poison is
governed by the poison Thiele modulus defined as

φp = R kp /De,p (4.1)
which represents the ratio between the rate of poison deposition (assumed to
be first-order with respect to poison concentration) and the rate of poison dif-
fusion inside the pellet. In this work, all diffusion coefficients were assumed to
be constant with respect to time, which is true as long as the poison loading is
not sufficiently large to significantly reduce the catalyst pore-mouth. Depending
on the value of the poison Thiele modulus, two characteristic cases are obtained.
For φp 1, the poison deposits uniformly throughout the pellet (uniform poison-
ing), while for φp 1 the poison deposits preferentially in a thin layer close to
the pellet’s external surface (pore-mouth poisoning). Four types of active element
distributions were considered: uniform, eggshell, egg-white, and egg-yolk. When
the main reaction is under diffusion control (large φ 0 ), as in the nondeactivating
case (section 2.1.1), the best performance for uniform poisoning is provided by
86
4.1 Nonselective Poisoning 87
Figure 4.1. Catalyst selection chart for maximization

of catalyst lifetime. (Active-catalyst volume)/(total sup-
port volume) = 1/3. First-order reaction with first-order
poison deposition. (From Becker and Wei, 1977b.)
an eggshell distribution. However, under pore-mouth poisoning conditions, a sig-

nificant improvement in catalyst performance can be achieved by distributing the
active element in the inner region of the pellet. These considerations lead to the
catalyst selection chart shown in Figure 4.1.
Similar improvement under pore-mouth poisoning was also demonstrated for
uniform catalysts covered with an inert protective layer composed of a different
support material, hence with different diffusion properties (Wolf, 1977; Polomski
and Wolf, 1978). For CO oxidation, this type of catalyst showed an enhancement
in activity when operating in the region of high reactant concentrations where the
reaction exhibits negative-order kinetics, whereas at lower concentrations, where
the reaction is essentially first-order, only the catalyst lifetime could be improved.
Bacaros et al. (1987) and Pavlou and Vayenas (1990b) addressed the prob-
lem of optimal initial activity distribution in isothermal pellets undergoing pore-
mouth poisoning. First-order kinetics were examined, and it was assumed that
catalyst poisoning was much slower than diffusion within the pellet, so that the
quasi–steady-state assumption could be employed in the mass balance of the main
reactant. The dimensionless model equations are

1 d n du
s = a(s)φ 2 u, 0 < s < sp (4.2a)
s n ds ds

1 d n du
s = 0, sp < s < 1 (4.2b)
s n ds ds
88 Studies Involving Catalyst Deactivation
du
s = 0: =0 (4.3a)
ds
s = 1: u=1 (4.3b)
where sp corresponds to the location of the deactivation front. Catalyst poisoning

was described by the Voorhies equation, which implies that the front sp advances
towards the pellet interior independently of the concentrations of the catalyst,
reactants, or poison:
sp = 1 − π (4.4)
where is dimensionless time and π is an empirical coefficient ranging from 0.5 to

1 (Carberry, 1976). The performance index used for optimization was the average
reactant conversion over the catalyst lifetime,
1
H= η() d. (4.5)
0
At low Thiele modulus (φ → 0), it was found that the optimal distribution is a
Dirac-delta located at the pellet center. This is because under kinetic control,
any catalyst distribution provides the same performance if no deactivation occurs,
while the time required for the poison to reach the catalyst is maximum if the
catalyst is located at the pellet center.
For large Thiele-modulus, the optimal initial activity distribution tends to be
bimodal, concentrating one portion of the active element near the pellet’s external
surface and another portion close to its center (Pavlou and Vayenas, 1990b), as
illustrated in Figure 4.2.a for π = 1. The specific shape depends on the pellet
geometry, values of the physicochemical parameters involved, and kinetics of the
deactivating process through the value of the parameter π . The catalyst loading
required near the center increases in going from the slab to the sphere geometry,
Figure 4.2. Optimal catalyst distribution for nonselective poisoning. First-order main re-
action with (a) constant velocity of the deactivating front (π = 1), (b) diffusion-controlled
deactivation (π = 0.5). Here φ[(n + 1)/gv ]1/2 = 20. (From Pavlou and Vayenas, 1990b.)
4.2 Selective Poisoning 89
with a concomitant decrease near the external surface. When π = 0.5, the catalyst
loading is substantially smaller near the surface and in fact vanishes at the surface,
as shown in Figure 4.2.b. This behavior results from the interaction of two opposing
factors. One is the effectiveness factor, which dictates that the catalyst should be
placed close to the surface. The other is the catalyst lifetime, which dictates that
it should be placed near the center. As seen in Figure 4.2.a, these factors result in
a bimodal optimal distribution for π = 1. However, when π = 0.5, the poisoning
front moves faster at the beginning. Consequently, the optimal catalyst loading
near the pellet surface decreases substantially. Finally, it should be noted that for
the case examined, contrary to nondeactivating systems, depositing the catalyst
nonuniformly does not improve the pellet performance significantly.
For an isothermal first-order reaction, the case of uniform poisoning, in both
single pellets and fixed-bed reactors, was considered by DeLancey (1973). The
objective function included economic considerations related to conversion of re-
actants and catalyst replacement. An approximate solution was derived using
Pontryagin’s principle. For a single pellet the optimal catalyst distribution was
eggshell, while for a fixed-bed reactor it was uniform.
4.2 Selective Poisoning

In contrast to the case of nonselective poisoning discussed above, where only
specific situations have been investigated, a result of general validity has been
obtained in the case of selective poisoning (Brunovska et al., 1990). In general,
the poisoning rate depends on the concentration of poison species, the concentra-
tions of reactants and products, and the temperature. Since the poisoning rate is
usually much lower than that of the main reaction, the quasi–steady-state approx-
imation can be adopted. In addition, it was assumed that the rate of deactivation
depends linearly on the local active catalyst concentration. The model equations
in dimensionless form are

1 d n du
s = a(s, t)φ 2 f (u, up , θ ) (4.6)
s n ds ds

1 d n du p
s = a(s, t)φp2 fp (u, up , θ ) (4.7)
s n ds ds

1 d n dθ
s = −a(s, t)βφ 2 f (u, up , θ ) (4.8)
s n ds ds
with BCs
du dup dθ
s = 0: = 0, = 0, =0 (4.9a)
ds ds ds
s = 1: u = 1, up = 1, θ = 1. (4.9b)
The deactivation is accounted for by a balance of active sites, which in terms of
the activity distribution function takes the form
∂a
= −a fp (u, up , θ ) (4.10)
∂
with initial condition
= 0: a = a(s, 0) (4.11)
An economic criterion based on profit per unit time was used as the performance
index, which takes into account price of product and cost of catalyst replacement:

α1 0 oper η() d − α3
G= (4.12)
oper
where oper is the dimensionless operating time and α1 , α3 are weighting coeffi-
cients proportional to product value and catalyst cost, respectively. The optimiza-
tion variables were the initial catalyst distribution a(s, 0) and the operating time.
By developing an optimality condition and following similar arguments to those
in section 2.3.2, it was shown that the optimal distribution is a Dirac-delta func-
tion if no upper bound is imposed on the local catalyst loading. In the case where
local loading is bounded, the optimal distribution is expected to take the form of a
step function, in analogy with nondeactivating systems described in section 2.5.3.
This result is valid for multiple reactions, following arbitrary kinetics, in the pres-
ence of thermal effects as well as inter- and intraparticle mass and heat transport
resistances.
The case of an isothermal first-order reaction where the poisoning kinetics is
also first-order was investigated as an example (Brunovska et al., 1990). It was
found that for low values of the ratio of the two Thiele moduli characteristic of
the poisoning and main reactions, φ p /φ, the optimal location is at the pellet exter-
nal surface, because the intraparticle resistance is larger for the reactant than for
the poison. For increasing values of this parameter, the optimal location moves
towards the pellet interior, to an extent which depends on the specific operating
conditions. Further, the optimal operating time increases monotonically with φp /φ.
The case of selective poisoning in a fixed-bed reactor was studied by Markos
and Brunovska (1989) for a nonisothermal plug-flow reactor. Optimization was
carried out only for Dirac-type distributions constituting a single-zone reactor, and
in all cases significant improvements were reported when using the appropriate
Dirac-delta distribution. First-order deactivation kinetics (on poison concentra-
tion) and first-order kinetics of the main reaction (on reactant concentration) were
considered. The objective function took account of the catalyst and raw-materials
cost as well as the value of the products. It was defined similarly to (4.12), but in
this case the integral of outlet reactor conversion was used

(α1 − α2 ) 0 oper X d − α3
K= (4.13)
oper
where α2 is a weighting coefficient proportional to the cost of raw materials.
Two competing factors determine the optimal catalyst location. Surface loca-
tions favor high initial reactor conversion, while internal locations are beneficial
for longer catalyst lifetime. This can be seen in Figure 4.3, where the objective
function is shown as a function of the operating time oper for various active
Figure 4.3. Time dependence of objec-

tive function given by equation (4.13)
for various active-layer locations; φ p =
10, Dap = 10, (a) (α1 − α2 )/α3 = 0.1,
(b) (α1 − α2 )/α3 = 10. Isothermal first-
order reaction, poisoning kinetics first-
order. (From Markos and Brunovska,
1989.)
catalyst locations, and cases of an expensive [i.e. α3 (α1 − α2 )] or cheap [i.e.

α3 (α1 − α2 )] catalyst. In the first case (Figure 4.3.a), the objective function is
maximized for s̄ = 0.3 and oper ∼ 80, while in the second (Figure 4.3.b), the op-
timal performance is reached for s̄ ∼ 0.6 and oper ∼ 20. Thus, when the catalyst
cost is high, it is better to locate the active element in the pellet interior and have
long operation time, while if the catalyst cost is low, the optimal location moves
towards the pellet surface in order to favor the main reactant transport process,
and the operating time decreases. The case of a single second-order reaction and of
two first-order consecutive reactions was also investigated numerically by Markos
et al. (1990), using an isothermal plug-flow heterogeneous reactor model.

In this section, we present results from a variety of experimental studies related to
the behavior of nonuniform catalyst distributions in the presence of deactivation.
These include methanation, olefin hydrogenation, and NO reduction reactions.
Catalyst design under poisoning conditions is critical in the context of automotive
exhaust catalysis, and is discussed in section 6.1.
4.3.1 Methanation
The methanation reaction
CO + 3H2 → CH4 + H2 O (4.14)

Figure 4.4. Effectiveness factor as a function of active-layer

location for step-type Ni/Al2 O3 pellets during CO methana-
tion in the presence of coking. The solid line represents the
simulation results, the circles denote experimental data, and
the dashed line shows the calculated CH2 /CCO ratio at the
active layer location. T = 573 K, CCO,f = 5 × 10−6 mol/cm3 ,
CH2 , f = 10.5 × 10−6 mol/cm3 . (From Wu et al., 1990b.)
over Ni/Al2 O3 was studied in step-type catalyst pellets by Wu et al. (1990b). This
reaction is accompanied by selective poisoning due to carbon deposition (coking)
when the H2 /CO ratio is low. The pellets were prepared by pressing an active pow-
der (Ni/Al2 O3 ) layer sandwiched between two inert powder (Al2 O3 ) layers. It was
found that significant coking occurred only when the active layer was located close
to the external surface of the pellet. By placing the catalyst layer deeper inside,
diffusional resistances introduced by the outer Al2 O3 layer increased the H2 /CO
ratio at the catalyst location, since the diffusivity of hydrogen is larger than that of
carbon monoxide in the Knudsen diffusion regime. High H2 /CO4 ratios suppress
coking, but if the catalyst is placed too deep inside the pellet, the effectiveness fac-
tor decreases significantly. Accordingly, for the reaction conditions investigated,
a pellet with the catalyst layer centered at s̄ ∼ 0.6 had the highest effectiveness
factor while simultaneously avoiding coking, as shown in Figure 4.4. Agreement
of experiments with theoretical diffusion–reaction calculations was good, as also
shown in the figure.
4.3.2 Hydrogenation
Ethylene hydrogenation in the presence of low concentrations of the poison thio-
phene on Pt/Al2 O3 catalysts was studied by Brunovska and coworkers (Pranda
and Brunovska, 1993; Remiarova et al., 1993; Brunovska et al., 1994). Pellets were
prepared both by impregnation of preformed support and by pressing active and

inert powder layers. The performance index used for comparison was the total
reactant conversion over the catalyst lifetime, equation (4.5).
Experiments demonstrated that the selective poisoning process consists of two
consecutive steps: reversible adsorption of poison precursor, followed by slow ir-
reversible surface reaction. Initial effectiveness factors were found to be larger
for the surface catalyst pellets, indicating that mass transfer resistances decreased
the hydrogenation rate. The maximum value of the performance index was found
to depend on the operating time. For short operating times, the surface catalyst
pellet showed slightly better performance, since deactivation was limited. For
longer times subsurface distributions were optimal, showing that mass transfer
resistance retarded pellet deactivation. In addition, the dependence of the per-
formance index on catalyst location was not monotonic. In some cases two max-
ima appeared, whose location and relative magnitude depended on the operating
time and width of the active layer. This behavior is due to the interaction between
complex main and poisoning reaction kinetics. While maintaining the center of the
active layer at a fixed subsurface location, the effect of layer width was also investi-
gated. As shown in Figure 4.5, the thinnest layer (i.e. approaching the Dirac-delta
distribution) exhibited the best performance, as expected from the theoretical
results presented in section 4.2.
Lin and Chou (1994) investigated the selective hydrogenation of isoprene on
Pd/γ -Al2 O3 catalysts, which is representative of partial hydrogenation of pyrol-
ysis gasoline from naptha cracking to increase its octane number. This is a con-
secutive reaction system with isoprene hydrogenating to isopentenes, which in
turn hydrogenate to isopentane. Coke buildup from the polymerization of di-
alkenes is the main mechanism for catalyst deactivation. Eggshell and uniform
catalysts were prepared by impregnation of the support with a toluene solution of
Pd(CH3 COO)2 and an aqueous solution of Pd(NH3 )4 (NO3 )2 , respectively. Both
◦
catalysts contained 0.2 wt% Pd with the same particle size of 8 A. The selectiv-
ity to isopentenes was higher for the eggshell catalyst, since they diffused away
from the reaction zone before hydrogenation to isopentane. The eggshell catalyst
Figure 4.5. Performance index H given by equation (4.5) as

a function of time for three step catalyst distributions: curve
a, s̄1 = 0.90, s̄2 = 0.93; curve b, s̄1 = 0.86, s̄2 = 0.97; curve c,
s̄1 = 0.83, s̄2 = 1.0. (From Pranda and Brunovska, 1993.)
also showed smaller coke formation and slower deactivation. This was attributed
to the higher hydrogen partial pressure in the reaction zone, which inhibits coke
buildup and further polymerization. In addition, coke precursors could diffuse out
from the catalyst pores more easily, hence decreasing coke formation. However,
palladium migration from the outside towards the core of the pellet occurred for
eggshell catalysts (Lin and Chou, 1995), during hydrogenation as well as subse-
quent catalyst regeneration. This migration was accompanied by sintering, which
resulted in loss of catalytic surface area.
4.3.3 NO Reduction
Stenger et al. (1988) and Hepburn et al., (1991b) studied the design of Rh/Al2 O3
honeycomb catalysts for NO reduction with H2 , in the presence of SO2 , for use in
stationary pollution sources such as power plants and industrial boilers. Eggshell
and egg-white catalysts were prepared by coimpregnation of the support with
RhCl3 and HF. Electron probe microanalysis revealed that sulfur was distributed
throughout the support, indicating that SO2 was not limited by diffusion. Thus,
subsurface location of Rh does not delay deactivation. It is worth noting that
sulfur deposition on alumina was small, while higher sulfur concentrations were
observed where Rh was present, indicating selective poisoning. The egg-white
catalyst exhibited greater tolerance to deactivation than the eggshell catalyst,
especially at longer periods. This behavior was attributed to the presence of the
outer alumina layer, which scavenges the SO2 before it reaches the noble metal.
5
Membrane Reactors
M
embrane reactors offer the advantage over conventional fixed-bed reactors
of combining chemical reaction and separation in a single unit. They can
substantially improve the performance of reactions by selective removal of
one of the reaction products or by controlled addition of a reactant. In the former
case, conversion enhancement beyond the thermodynamic limit can be achieved
for equilibrium-limited reactions, while in the latter, product selectivity can be
improved by influencing the concentration and residence time of components
giving rise to undesired reactions. Several articles which review experimental and
theoretical studies of catalytic membrane reactors are available in the literature
(Hsieh, 1991; Tsotsis et al., 1993; Saracco and Speccia, 1994; Zaman and Chakma,
1994).
5.1 Membrane Reactors with Nonuniform Catalyst Distribution

In this section we discuss theoretical studies addressing nonuniform distribution
of catalyst in membrane reactors, in order to gain insight into the effect of cat-
alyst location on this new reactor type, before addressing optimization issues in
the following sections. Yeung et al. (1994) investigated the influence of the loca-
tion of a Dirac-delta catalyst in pellets contained in an inert membrane reactor
with catalyst on the feed side (IMRCF), a catalytic membrane reactor (CMR),
and a conventional fixed-bed reactor (FBR). For the IMRCF and the FBR, the
active catalyst is distributed in pellets placed inside the membrane or the reactor
tube respectively, while for the CMR the active catalyst is distributed within the
membrane itself as shown in Figure 5.1.
In order to provide analytical expressions so that the effects of the use of a
membrane as well as nonuniform catalyst distribution are readily identified, an
isothermal, first-order reversible reaction A B was considered, under well-
mixed fluid-phase conditions and with no interphase mass transfer resistances.
Furthermore, pure inert and pure A are fed to the permeate and feed sides, respec-
tively of the IMRCF and CMR. The function of the inert is to sweep component B
produced in the reactor and diffusing towards the permeate side. The steady-state
95
96 Membrane Reactors
Figure 5.1. Schematic diagrams of (a) inert membrane reactor with

catalyst on the feed side (IMRCF), (b) catalytic membrane reactor
(CMR), and (c) fixed-bed reactor (FBR). (From Yeung et al., 1994.)
mass balances for the three reactors were solved analytically for uniform and Dirac
catalyst distributions (see Szegner, 1997 for details). For the membrane reactors,
the total conversion XT is computed by combining the feed and permeate sides.
It is thus defined as the ratio between the total molar flow of reactant converted
to product and the initial molar flow of reactant on the feed and permeate sides:
ζF,A
0
+ ζP,A
0
"p0 − ζF,A "F − ζP,A "p
XT = (5.1)
ζF,A
0
+ ζP,A
0
"P0
where ζF,A , ζP,A , "F , and "P are the reactant mole fractions and dimensionless
molar flow rates on the feed and permeate sides, respectively, with
QF QP
"F = , "P = . (5.2)
QF0 QF0
The dimensionless residence time is defined as the ratio between residence time
and the characteristic time for diffusion:
Vr PF /Rg T QF0 De,A Vr PF
= = . (5.3)
δM /De,A
2
Rg T QF0 δM
2
Figure 5.2. The effect of location of a Dirac-delta catalyst

activity distribution on the total conversion XT as a function
of dimensionless residence time for an isothermal first-
order reversible reaction A B, for well-mixed (a) IMRCF
and (b) CMR. (From Yeung et al., 1994.)
The variation of the total conversion as a function of residence time is shown

in Figure 5.2 for the two membrane reactors as the location of the Dirac catalyst is
varied in the catalyst pellets (for the IMRCF) and in the catalytic membrane (for
the CMR). In each case, the performance of the corresponding uniform catalyst
distribution is also shown for comparison. For the IMRCF (Figure 5.2.a), it is
clearly seen that on moving the catalyst from a position near the center (s̄ = 0.1)
to the surface (s̄ = 1) of the pellets, the conversion XT increases to larger values
than in the uniformly distributed case [a(s) = 1]. This occurs because of decreased
intraphase mass transfer resistance as the Dirac distribution is located closer to
the pellet surface. A similar result is observed for the CMR (Figure 5.2.b) as
the Dirac-delta distribution is moved from the permeate side (ξ̄ = 0) to the feed
side of the membrane (ξ̄ = 1). When the catalyst is placed on the permeate side,
the conversion does not exceed the equilibrium value, because in this case the
Figure 5.3. Total conversion XT for well-mixed IMRCF,

CMR, and FBR with uniform and Dirac-delta catalyst ac-
tivity distributions, as a function of dimensionless residence
time , for an isothermal first-order reversible reaction
A B. (From Yeung et al., 1994.)
selective product separation does not play a role in enhancing the conversion. On
comparing results of the two membranes (i.e. Figure 5.2.a and b) it is seen that the
improvement in reactor performance on locating the Dirac catalyst at s̄ or ξ̄ = 1, as
compared to the uniform catalyst distribution, is greater for the CMR. This occurs
because in the case of the uniform CMR, as the reactant diffuses through the
membrane, the back reaction of the product competes with its selective removal.
In the case of the IMRCF, the difference in the performance between the uniform
and the surface distribution is not as large, because the separation occurs outside
the reaction zone.
A comparison of the various reactor configurations is shown in Figure 5.3.
The function of the membrane in this system is evident on considering that at
large residence time the fixed-bed performance is limited to XT = 0.5 by chemical
equilibrium. That limit is exceeded by all membrane reactors. The Dirac-delta dis-
tribution located at the feed side or at the pellet surface (i.e. ξ̄ = 1 for the CMR
and s̄ = 1 for the IMRCF and FBR) shows superior performance to a uniform
activity profile. This distribution maximizes the reactant concentration at the cat-
alyst site, since it is not limited by diffusion through the membrane or the pellet.
The performance of the CMR and IMRCF is identical when the Dirac location
is at the feed side of the membrane (for the CMR) or at the external surface
of the pellet (for the IMRCF), due to the absence of mass transfer resistances.
These two membrane configurations exceed the performance of the FBR con-
figurations over the full range of residence time. However, for small the FBR
with s̄ = 1 (i.e. Dirac-delta at the external surface of the pellet) performs better
than the IMRCF with a(s) = 1 (i.e. uniform catalyst distribution). This is because
at small residence time, due to diffusional limitations, the reactant sees only a
Figure 5.4. A comparison of well-mixed IMRCF, CMR, and FBR with uniform [a(s) = 1 or
a(ξ ) = 1] and Dirac-delta (at s̄ = 1 or ξ̄ = 1) catalyst activity distributions for (a) product
purity on the feed side ζF,B , (b) product purity on the permeate side ζP,B , (c) product
molar flow rate on the feed side "F ζF,B , and (d) product molar flowrate on the permeate
side "P ζP,B , as a function of dimensionless residence time . The system is an isothermal
first-order reversible reaction A B. (From Yeung et al., 1994.)
small amount of the total catalyst in the pellets of the IMRCF, but it is exposed
to the full amount of catalyst in the FBR, thus resulting in higher conversion to
product.
Similar arguments hold when product purity or product flow rate on either the
feed or the permeate side is used as the performance index. The purities on the
feed and permeate sides are the product mole fractions ζF,B and ζP,B respectively,
while the product flow rates are "F ζF,B and "P ζP,B . A comparison for these cases
is illustrated in Figure 5.4. It can be seen (Figure 5.4.a) that on the feed side
both FBRs perform better than the membrane reactors for most cases. This is
because the feed side is depleted of the products as a result of diffusion across
the membrane. This point is illustrated in Figure 5.4.c, where a maximum may be
observed. For small residence times, the product molar flow rate is small because
of insufficient contact with the catalyst, whereas for larger residence times the
product permeates through the membrane until total permeation occurs (i.e. no
flow out from the feed side). When the product purity or its flow rate on the
permeate side is to be maximized, it is seen in Figure 5.4.b and d that for both the
CMR and IMRCF a Dirac-delta distribution of catalyst located on the feed side
(i.e. ξ̄ = 1 for the CMR and s̄ = 1 for the IMRCF) shows better performance than
a uniform distribution.
5.2 Optimal Catalyst Distribution in Pellets for an Inert

Membrane Reactor
The problem of determining the optimal distribution of catalyst in pellets for
an inert membrane reactor with catalyst on the feed side (IMRCF), with well-
mixed fluid phases on both the feed and permeate sides, where multiple reactions
with arbitrary kinetics take place under nonisothermal conditions and external
transport resistances, has been solved (Yeung et al., 1994). The methodology is
similar to that followed previously in section 2.5.1. In this case, in addition to
the mass and energy balance equations for the pellet, fluid-phase balances for
the feed and permeate sides are also required. The resulting set of equations
is cumbersome and is presented along with the formulation of the optimization
problem in Appendix B.
Due to the complexity and nonlinearity of the equations involved, the solution
of the optimization problem can be challenging even using advanced numerical
techniques. In the general case, direct numerical methods must be employed (cf.
Baratti et al., 1993). However, when B equals zero (i.e. linear dependence of
catalytic surface area on loading), the Hamiltonian becomes a linear function
of the active element distribution µ(s), and then, as discussed in section 2.5.3,
an analytical solution can be obtained. In this case, the optimal distribution is a
multiple-step function, where the number of steps is controlled by the number of
sign changes of the switching function. This means that the presence of an inert
membrane and a permeate flow do not influence the nature of the optimal catalyst
distribution, although they may well alter the specific locations and the number
of steps. One would also expect that as the upper bound of local catalyst loading
tends to infinity, the step distribution would approach a Dirac-delta distribution.
It should be clear that the similarity of the above optimization solution to that of
a single pellet (section 2.5.3) exists because fluid phase conditions in the membrane
reactor are well mixed. If the fluid phases are not well mixed, then one can expect
that the optimal catalyst distribution in pellets will change with axial position in
the reactor, as in the fixed-bed reactor in section 3.1. However, a rigorous proof
of this conjecture is not yet available.
5.3 Optimal Catalyst Distribution in a Catalytic Membrane Reactor

The general problem of optimal catalyst distribution in a CMR has been studied
by Szegner (1997). The problem formulation is similar to that for the IMRCF (see
Appendix B), except that the catalyst is placed in the membrane itself and not in
the pellets. Thus again, in general, a numerical approach is required to determine
the optimal distribution that maximizes the Hamiltonian. In the case of B = 0, a
multiple-step distribution is proven to be optimal.
5.3 Optimal Catalyst Distribution in a Catalytic Membrane Reactor 101
The particular case of a planar catalytic membrane where the downstream bulk
flow is very large has been considered by Keller et al. (1984). In this case we can
assume that the downstream concentration of the reactant diffusing through the
membrane is zero. Under isothermal conditions and in the absence of external
transport resistances, the problem reduces to the diffusion–reaction equation in
the catalytic membrane:
d2 u
= φ 2 a(s) f (u) (5.4)
ds 2
together with the boundary conditions
s = 0: u=1 (5.5a)
s = 1: u = 0. (5.5b)
Note that this problem differs from that of a catalyst pellet, given by equations
(2.63)–(2.66), only in the BCs. The space coordinate of the membrane in this
case starts at the side in contact with the feed stream (s = 0) and finishes at the
downstream side (s = 1).
For reaction rates f (u) that increase monotonically with reactant concentra-
tion, and with effectiveness factor as the performance index, a Dirac-delta function
located at the feed side surface of the membrane is found to be the optimal catalyst
distribution, in accordance with the analysis in section 2.1. This is where the high-
est reactant concentration is encountered, which is beneficial for a positive-order
reaction.
A different answer for the optimization problem is obtained when minimizing
the flux of unreacted material leaving the membrane,

du
I= − . (5.6)
ds s=1
Using a specific variational analysis applied to the case of linear kinetics, i.e. f (u) =
u, Keller et al. (1984) found that the optimal activity distribution is given by the
following step function:

1/L, s2 ≤ s ≤ s2 + L
a(s) = (5.7)
0, 0 ≤ s < s2 and s2 + L < s ≤ 1
where
√
L
s2 = 1 − L − (5.8a)
φ

1 − 1 + φ2
L= 1+2 . (5.8b)
φ2
Even though both the objective functional (5.6) and the BCs (5.5) are not the
same as in the case of the catalyst pellet, it is surprising that here we encounter for
the first time an optimal activity distribution which is not a Dirac-delta function.
In order to understand this point, following the procedure described in section
2.3.1, we can derive the necessary condition for optimality,

1 1
∗ ∗
G a (s) ds ≥ G ∗ a(s) ds (5.9)
0 0
which is the same as (2.81) with n = 0, except that the function G ∗ is now given by
G ∗ = φ 2 λu ∗ (5.10)
and the Lagrange multiplier λ satisfies
d2 λ
= φ2a∗λ (5.11)
ds 2
with BCs
s = 0: λ=0 (5.12a)
s = 1: λ = 1. (5.12b)
By taking (5.7) as the optimal distribution a ∗ and substituting in equations (5.10)–
(5.12), we obtain
 √

 φ 2 s(1 − s − L)(1 − L)−2 exp(−φ L), 0 ≤ s < s2
 √
∗ −1
G = φ [1 − s2 (1 − L) ] exp(−φ L),
2 2
s2 ≤ s ≤ s2 + L . (5.13)

φ 2 (1 − s)(s − L)(1 − L)−2 exp(−φ √ L), s + L < s ≤ 1

2
It can be seen that in the interval s2 ≤ s ≤ s2 + L, where the membrane is catalyt-

ically active (i.e. a ∗ = 0), the function G ∗ (s) is constant and attains its maximum
value. Therefore, we are in the peculiar situation discussed in the context of equa-
tion (2.89), where the necessary condition (5.9) cannot be violated by selecting
an activity distribution a(s) with the shape of a Dirac-delta function. This is in
agreement with the result of Keller et al. (1984), who conclude that the optimal
distribution is a step and not a Dirac-delta function.

Because the topic of optimal catalyst distribution in membrane reactors started re-
ceiving attention only recently, few experimental studies have been reported. The
first set consists of dehydrogenation reactions in packed-bed membrane reactors,
where the catalyst distribution has been shown to influence reactor performance.
Secondly, preparation of catalytic membranes with controlled active layer thick-
ness and location is addressed.
5.4.1 Dehydrogenation Reactions

Cannon and Hacskaylo (1992) demonstrated that catalyst location and amount
in the membrane can affect reactor performance. They investigated cyclohexane
dehydrogenation in porous Vycor glass tubes packed with 0.5-wt% Pd/Al2 O3 cat-
alyst pellets. Three membranes were used: (a) a pure Vycor glass membrane, (b) a
Figure 5.5. Effect of residence time,

W/Q, on the conversion of ethane in
an IMRCF with different catalyst dis-
tributions during ethane dehydrogena-
tion on Pt-Sn/γ -Al2 O3 pellets. Feed: 20%
C2 H6 /Ar. (From Yeung et al., 1996.)
Vycor glass membrane containing 0.25% Pd, which was located preferentially
at the inner and outer surfaces of the glass tube (PM1), and (c) a Vycor glass
membrane containing 0.5% Pd, which was distributed uniformly throughout the
membrane wall (PM2). In all cases, total cyclohexane conversion increased with
increasing sweep-gas flow rate on the permeate side, due to selective and con-
tinuous removal of hydrogen from the reaction zone. In general, the presence of
palladium particles in the membrane decreased cyclohexane conversion owing to
the reverse reaction, and this effect was larger for PM2 than for PM1.
Yeung et al. (1996) studied ethane dehydrogenation to ethylene in an IMRCF
using Pt-Sn/γ -Al2 O3 catalysts. Three types of pellets were prepared with the same
total metal loadings (1.1 wt% Pt, 1.3 wt% Sn): (a) uniform Pt distribution, (b) nar-
row surface-step Pt distribution (Pt layer thickness ∼10% of pellet radius), and
(c) wide surface-step Pt distribution (Pt layer thickness ∼40% of pellet radius).
In all cases, Sn (added to enhance ethylene selectivity) was distributed uniformly
in the pellets. The membrane was a thin dense film of palladium deposited on
a porous 316 stainless-steel tube that permits transport only of hydrogen. As
shown in Figure 5.5, ethane conversion increases with residence time for both
nonuniform distributions, and is two to three times higher than the equilibrium
conversion. The highest conversion is attained by concentrating the Pt catalyst
at the pellet external surface. This indicates that the influence of transport pro-
cesses in the pellet is not negligible and their presence is detrimental to catalyst
performance.
Szegner et al. (1997) studied experimentally and theoretically the same reac-
tion on step-type Pt-Sn/γ -Al2 O3 catalysts in a well-mixed isothermal IMRCF.
Four types of pellets with different Pt distributions but with approximately the
same amount of platinum and tin were used, as shown in Figure 5.6. The membrane
used was a composite mesoporous γ -Al2 O3 membrane. In order to minimize reac-
tant loss, the effective permeability of the commercial composite MembraloxTM
membrane with 2-µm γ -Al2 O3 membrane layer was reduced by slip-casting of
alumina sol. The slip-cast modified membrane was approximately 17 µm thick
(see Figure 5.7 for cross sections of original and slip-cast modified Membralox
membranes). As shown in Figure 5.8, optimum reactor performance was ob-
tained when the catalyst was concentrated at the surface of the pellet. In this case
Figure 5.6. Pt-Sn/γ -Al2 O3 pellets with different platinum distributions:
Active-layer Active-layer Pt loading Sn loading

Part Catalyst distribution center s̄ width 2 (wt%) (wt%)
a Narrow surface step 0.95 0.10 1.11 1.36
b Wide surface step 0.875 0.25 1.17 1.43
c Near surface step 0.70 0.10 1.09 1.34
d Deep subsurface step 0.50 0.10 1.16 1.35
(From Szegner et al., 1997.)
supraequilibrium conversions, about 80% beyond equilibrium values, were ob-

served. Ethane conversion decreased as the catalyst location moved from the
surface towards the pellet interior. This is due to strong diffusional resistance in
the inert region before the reactant contacts the active sites within the catalyst
pellets. Since dehydrogenation of ethane is a positive-order reaction, decreasing
reaction rates are obtained as a consequence of the reactant concentration gra-
dient. Hence, by placing the active layer as a narrow step at the pellet surface,
intrapellet diffusion limitations are minimized, leading to higher reaction rates. It
is worth noting that, owing to high intrinsic catalyst activity, conversions obtained
using wide-surface-step pellets (Figure 5.6.b) were only slightly lower than with
narrow-surface-step pellets (Figure 5.6.a). The performance of the IMRCF con-
taining pellets with the deepest catalyst location (Figure 5.6.d) was inferior even
to that of a fixed-bed reactor packed with surface-catalyst pellets. The theoretical
model assumed isothermal, steady-state, well-mixed conditions on both sides of
the membrane, and included diffusion–reaction in the pellets and Knudsen diffu-
sion across the membrane. The agreement between numerical and experimental
Figure 5.7. SEM micrographs of Membralox tubular mem-

branes: (a) original and (b) slip-cast modified. (From Szegner
et al., 1997.)
Figure 5.8. Effect of temperature on conversion of

ethane in an IMRCF with platinum distributed as
a narrow surface step (solid curve), near-surface
step (long-dashed curve), and deep subsurface step
(medium-dashed curve), and in an FBR with plat-
inum distributed as a narrow surface step (short-
dashed curve). The symbols and lines are the exper-
imental data and model results, respectively. (From
Szegner et al., 1997.)
results was good (see Figure 5.8), especially considering that no adjustable param-
eters were used, since intrinsic reaction kinetics and diffusivities were determined
from separate experiments.
5.4.2 Preparation of Catalytic Membranes

For the preparation of catalytic membranes with nonuniform distribution of cat-
alyst, conventional impregnation methods discussed in Chapter 7 cannot be ap-
plied, because the typical membrane thickness is too small (2–50 µm). In addition,
it is difficult to prevent the penetration of catalyst precursor into the graded sup-
port material, which is undesirable. For example, in the impregnation of graded-
structure alumina membranes (Membralox) with platinum precursors, it was ob-
served that while most of the metal was deposited uniformly on the top γ -Al2 O3
layer, rich in hydroxyl groups, part of it also penetrated the α-Al2 O3 support (cf.
Uzio et al., 1991).
Crack-free alumina membranes with nonuniform distribution of catalyst can
be prepared by sequential slip-casting (Yeung et al., 1994, 1997). According to
this technique, the membrane is built layer by layer, alternately slip-casting inert
alumina and catalyst-containing sols to achieve the desired active layer width and
location within the membrane. Some examples of catalytic membranes containing
Pt are shown in Figure 5.9. Figure 5.9.a shows a 6-µm-thick γ -Al2 O3 membrane,
with a narrow 0.6-µm active layer, consisting of a 5-wt% Pt/alumina layer located
1.5 µm from the membrane surface. Figure 5.9.b is a micrograph of a 7.2-µm-thick
Figure 5.9. Nonuniform distribution of

platinum catalyst in membranes pre-
pared by the sequential slip-casting
technique: 1, 5-wt% Pt/γ -alumina 2,
γ -alumina 3, intermediate aluminas.
(From Yeung et al., 1997.)
γ -Al2 O3 membrane, with 1.5-µm-thick 5-wt% Pt/alumina located 3.3 µm from

◦
the membrane surface. The membrane top layer has a nominal pore size of 50 A.
Region 3 in both figures shows graded layers of a Membralox ceramic tube. These
scanning electron micrographs show that precise control of the active-layer width
and location can be attained using this technique. In addition, this preparation
method restricts the catalyst to the active layer, preventing catalyst loss onto the
graded ceramic support.
As mentioned in section 2.4.2, the relation of catalytic surface area vs loading
is critical for determining the optimal distribution of catalyst, and this relation can
be affected by the preparation conditions of the active sol. The active sol used
for the catalytic membranes shown in Figure 5.9 is prepared by ion-exchanging
chloroplatinic acid with a colloidal suspension of alumina (A1 sol: 25 wt% alu-
mina, average particle size 150 nm; A2 sol: 20 wt% alumina, average particle size
50 nm). By maintaining the pH at 4 during ion exchange, a linear relationship
between surface area and loading is obtained as shown in Figure 5.10. In this case,
Pt dispersions are high, as shown in Table 5.1. If buffer is not used, and the pH
is allowed to decrease, the relationship between surface area and loading is non-
linear, exhibiting an upper threshold of 0.75 m2 /g for Pt loading >1 wt%, and a
consequent loss of dispersion. These results indicate that the presence of buffer
leads to higher dispersion and a smaller Pt crystallite size.
Table 5.1. Platinum catalyst surface area, average crystallite size,

and dispersion.
Pt loading Pt surface area Pt crystallite size

(wt%) (m2 /g) (nm) Dispersion
Pt-A2 (pH = 4)
0.1 0.13 2.12 0.53
0.5 0.58 2.43 0.47
1.0 0.51 5.50 0.21
1.7 0.97 4.92 0.23
2.5 1.69 4.13 0.27
5.0 5.02 4.64 0.24
Pt-A2 (no buffer)
0.1 0.05 5.43 0.21
0.5 0.29 4.89 0.23
1.0 0.68 4.10 0.28
2.6 0.70 10.54 0.11
5.1 0.75 18.87 0.06
Different distributions of catalyst within the membrane can be obtained by

varying the thickness of the slip-cast layers (both inert and active) as well as
their arrangement. The widths of the layers are controlled by the slip-casting
parameters, i.e. slip-casting time, viscosity, slip concentration, particle size, alu-
mina concentration and support properties (i.e. pore structure, wettability). More
specifically, the thicknesses of the inert (A1) and active (5-wt% Pt-A1) membrane
layers vary linearly with the square root of slip-casting time as shown in Figure
5.11.a. This is consistent with the equation that Adcock and McDowall (1957)
derived for the gel layer thickness Lg as a function of the system physicochemical
parameters, following the Kozeny–Carman development for filter pressing:
0.5
2Kg Pg t
Lg = (5.14)
η αg
where Kg is the permeability of the gel layer, Pg is the pressure drop across the
Figure 5.10. Platinum surface area as a func-

tion of platinum loading for a Pt-A2 sol:
curve a, buffer pH = 4; curve b, no buffer.
Figure 5.11. Membrane layer thickness as a function of slip-cast parameters for inert
2-wt% Al2 O3 and active 5-wt% Pt/alumina –2-wt% Al2 O3 sols. (a) Parameter: (time)1/2 ;
inert and active A1 sol. (b) Parameter: number of slip-cast layers; inert and active A1 sol,
(c) parameter: (time)1/2 ; inert and active A2 sol. (From Yeung et al., 1997.)
gel layer, t is time, η is the solvent viscosity, and αg is a constant dependent on both
the sol concentration and the support characteristics. Besides increasing the slip-
casting time, the thickness of the membrane can also be increased by multiple
slip-casting, as shown in Figure 5.11.b. In this experiment, it is observed that the
inert and active sols display different behavior after the first slip-cast, because
the support on which subsequent layers are deposited is different in the two cases.
In Figure 5.11.c, the dependence of layer thickness on time is shown for inert (A2)
and active (Pt-A2) sols. The A2 sols exhibit different behavior from the A1 sols
(compare Figure 5.11.a and c). For the same slip-casting time, sol A1, which has
larger alumina particles, gives a thicker layer than A2. However, the active layer
Pt-A2 is thicker than Pt-A1.
Figure 5.12.a shows the dependence of layer thickness (slip-casting time 5 s)
on the alumina concentration of the slip. The thicknesses for both A1 and A2
sols increase with alumina concentration, reaching a maximum at about 15 wt%.
The viscosity of the slip also increases with alumina content as shown in Figure
5.12.b. Both sols display sharp increases in viscosity for alumina concentration
Figure 5.12. (a) Membrane layer thickness

and (b) viscosity of alumina slips (A1 and
A2 sols) as a function of alumina content.
above 10 wt%, which results in the thickness leveling off as observed in Figure
5.12.a. Sol viscosity is important in preventing pore clogging as well as controlling
gelation rate of the slip-cast (Cot et al., 1988). The viscosity can also be controlled
by addition of binders such as PVA, which help regulate the gelling rate and reduce
crack formation in the final membrane.
6
Special Topics of Commercial Importance
I
n this chapter, we consider specific topics of significant commercial value where
nonuniform distribution plays an important role in catalyst design. These in-
clude catalysts for automotive exhaust cleanup, petroleum refining operations
such as hydrotreating and cracking, biotechnology, and acid catalysis. Particularly
in the case of automotive catalysis, nonuniform distribution of noble metals pro-
vides critical advantages for pollution abatement reactions and has been employed
extensively.
6.1 Automotive Exhaust Catalysts

Automobile exhaust is considered the main source of air pollution in urban areas.
The major pollutants in exhaust gas are carbon monoxide, hydrocarbons, and ox-
ides of nitrogen. Following the Federal Clean Air Act of 1970, which called for a
drastic reduction in these emissions, on all cars made in the United States since
1975, catalysts have been used to convert the pollutants into harmless gases. The
catalysts in pre-1981 automobiles were oxidation catalysts, which controlled car-
bon monoxide and hydrocarbons only, by oxidizing them to form carbon dioxide
and water. Their active components were platinum and palladium, which were de-
posited on substrates with a large surface area, either a monolith or pellets packed
in shallow, pancake-shaped converters. Starting with the 1981 model year, because
of stricter nitrogen oxide controls and fuel economy requirements, oxidation cat-
alysts were replaced by three-way catalysts (TWCs), which simultaneously control
all three of the major pollutants. They oxidize carbon monoxide and hydrocarbons
while reducing nitrogen oxides as well. To perform these tasks, commercial TWCs
contain platinum, palladium, rhodium, and cerium oxide as active components.
Platinum and palladium provide activity for carbon monoxide and hydrocarbon
oxidation, while rhodium is excellent for NOx reduction. Since there are opposing
requirements for oxidation and reduction catalytic reactions, the engine must op-
erate in a narrow window around the stoichiometric air–fuel ratio where all three
pollutants are removed efficiently (see Figure 6.1). Excellent reviews of the use
110
Figure 6.1. Steady-state conversion of carbon monoxide

(CO), hydrocarbons (HC), and nitrogen oxides (NOx )
for a three-way catalyst as a function of the air–fuel ratio.
(From Kummer, 1980.)
of catalysts in automotive exhausts are available in the literature (Kummer, 1980;

Taylor, 1984; Heck and Farrauto, 1995).
6.1.1 Design of Layered Catalysts

Under normal operating conditions the catalyst is at sufficiently high temperature
so that reactions are diffusion-controlled, and for this reason eggshell distribution
is beneficial. However, a complication arises from the occurrence of deactivation
processes: poisoning due to the presence of trace quantities of phosphorus, sulfur,
and metals (Pb, Zn, etc.) in the engine oil and the fuel, and thermal sintering.
Phosphorus and metal poisoning are essentially irreversible, while poisoning by
sulfur compounds is reversible (Summers and Hegedus, 1979). In addition, the
catalyst has to be resistant to frequent thermal cycles as well as mechanical vibra-
tions. Since the size of the automotive catalyst market is large (see section 1.1) and
the noble metals are expensive, optimization of their use is of particular interest.
Some of the earliest applications of subsurface step-type catalysts were for
automotive exhaust emission control. Michalko (1966a,b) and Hoekstra (1968)
patented noble-metal catalysts with subsurface location of the active components,
which showed improved long-term stability by delaying catalyst poisoning. The
poisoning process is nonselective, i.e., it involves both the support and the ac-
tive catalyst. Furthermore, for phosphorus and metal poisoning it is diffusion-
controlled, i.e., the poison penetrates the catalyst support in the form of a sharp
front, whose velocity is independent of the local concentration of the active el-
ements (Hegedus and Summers, 1977). Under these conditions, the support can
112 Special Topics of Commercial Importance
Figure 6.2. Hydrocarbon conversion as a function of time,

during accelerated poisoning experiments with pellets hav-
ing different catalyst distributions. Pt/Pd: Pt(exterior)/
Pd(interior); Pd/Pt: Pd(exterior)/Pt(interior); Pt–Pd: Pt
and Pd (both exterior); Pt: Pt(exterior); Pd: Pd(exterior).
(From Summers and Hegedus, 1978.)
be used as a poison-getter to protect the unpoisoned noble metals beyond the

poisoned zone.
Summers and Hegedus (1978) studied the case where only the two oxidizing
reactions occur, and compared several distributions of Pt and Pd: Pt(exterior)/Pd
(interior), Pd(exterior)/Pt(interior), Pt–Pd(exterior), Pt(exterior), Pd(exterior).
Subsurface impregnation of both Pt and Pd would offer the best protection from
poisoning, but was not considered due to increased mass transfer resistances. Hy-
drocarbon conversions during accelerated poisoning experiments for the various
distributions are shown in Figure 6.2. The catalyst pellet which was impregnated
with Pt at the outer shell and with Pd in a separate subsurface band (Pt/Pd) exhib-
ited the best performance before, as well as after, poisoning. This was attributed
to the different poison characteristics of Pt and Pd. Pt is more resistant to poison-
ing, with a nonzero residual activity, while Pd completely loses its activity once
the poisoned shell penetrates through the Pd-impregnated shell. The Pt–Pd coim-
pregnated catalyst (Pt–Pd) showed unsatisfactory performance because the two
elements form an alloy with Pd on its outer surface, so that coimpregnated Pt–Pd
catalysts tend to poison with the characteristics of Pd. For CO oxidation, the fresh
Pt–Pd catalyst showed the highest activity, but after accelerated sintering or poi-
soning experiments the Pt/Pd catalyst exhibited the best performance, because Pd
was protected from poisoning by subsurface impregnation.
The above multilayer design was extended to three-way catalysts by Hegedus
et al. (1979a) (see also Hegedus and Gumbleton, 1980). These catalysts included
Rh for NOx reduction, and its high cost compared to that of Pt and Pd, as well as
Figure 6.3. Multilayer catalyst design concept.
its strong sensitivity to phosphorus and lead poisoning, suggests using it in small
quantities with protection from poisoning. Based on these considerations, in each
pellet the noble metals were located in three separate layers, one underneath the
other, as illustrated in Figure 6.3. The first, at the pellet’s external surface, was
the Pt layer whose depth was determined so as to entirely contain the expected
penetration depth of the poison during the desired catalyst lifetime. Next was
located the Rh layer, so that diffusional resistances for NOx reduction would be
minimized and Rh would also be protected from poisoning. Finally, the Pd layer
was placed at the deepest location, providing good lightoff performance, especially
for aged catalysts, because of its good thermal stability. This configuration also
prevented undesirable alloy formation among the noble metals. While maintaining
the same noble metals and their order in the layer configuration, Vayenas and
coworkers (Vayenas et al., 1994; Papadakis et al., 1996) have proposed the use
of different supports that further enhance catalytic activity. Multilayer three-way
catalyst design remains an active area of research, and numerous publications
continue to appear in the patent and scientific literature.
There is a strong motivation to avoid using Rh due to its high cost, and hence
research has been directed towards alternative catalyst formulations. However,
Pd- and Pt-based catalysts are typically not as effective for NO conversion. Some
researchers have tried to overcome these disadvantages by using various additives.
Hu et al. (1996) improved the performance of Pd-only three-way catalysts using
a two-layer design, where the first layer contained Pd and the second Pd and
ceria. Both layers utilized alumina as the support and contained base-metal oxide
stabilizers. The Pd layer provided high activity at low temperatures, while the Pd–
Ce layer offered large oxygen storage capacity, which gave improved three-way
activity at high temperatures. The two-layer Pd catalyst exhibited performance
comparable to a Pt/Rh catalyst.
Automotive exhaust catalysts constitute a remarkable example of catalyst de-
sign despite the fact that in many instances, because of complexity of the problem
and lack of detailed expressions for the kinetics of the various reactions involved,
the design was based on experimental observations rather than on quantitative
models.
6.1.2 Nonuniform Axial Catalyst Distribution

It has been experimentally demonstrated by various investigators (cf. Komiyama
and Muraki, 1990; Beck et al., 1997) that poisons tend to deposit preferentially
near the monolith inlet. Hence, two-zone monolith configurations have been
suggested, where the upstream portion of the monolith is left unimpregnated.

Oh and Cavendish (1983) investigated theoretically the poison resistance of such
monoliths and found that the warm-up time of a poisoned monolith was lowest
for an intermediate length of the inert first zone. This was due to balance between
insufficient protection of the noble metal at short inert-zone lengths and long time
to reach reaction temperature at long lengths.
Two-zone monoliths have been shown to offer improvements in other respects
as well. Wu and Hammerle (1983) studied two-zone converters where the first zone
contained only Pd, while the second zone contained Pt and Rh. This combination
showed improved thermal resistance and lightoff performance while offering a
reduction in precious-metal cost. The reason was that activity of the second zone
could be protected from thermal damage by the presence of the first zone, as
long as the latter had sufficient thermal resistance to maintain some activity after
repeated high-temperature exposure. Accordingly, Pd was placed in the first zone
because it is more resistant to sintering under oxidizing conditions than Pt and Rh.
Another area where nonuniform catalyst distribution can be beneficial is reduc-
tion of cold-start emissions. In the first 1–2 min after cold start of an automobile,
the catalyst temperature is too low for the reactions to take place. As a result,
significant amounts of pollutants pass unconverted through the catalyst. They
constitute a large fraction of the total tailpipe emissions. Nonuniform catalyst
distribution along the monolith length can result in shorter warm-up period and
thus lower cold-start emissions. Oh and Cavendish (1982) examined the lightoff
behavior of three axial catalyst distributions: uniform, linear increasing, and linear
decreasing. They found that CO cold-start emissions are reduced when the noble
metal is concentrated in the upstream section of the monolith, as shown in Figure
6.4. If strong poisoning is present, this beneficial effect is counterbalanced by the
preferential poison accumulation in this region, so that it may be preferable to con-
centrate the noble metal in the downstream section of the monolith where poison
Figure 6.4. Effect of axial catalyst activity profile on

lightoff. (From Oh and Cavendish, 1982.)
Figure 6.5. Two-dimensional catalyst distribution con-

cept. (From Naoki et al., 1996.)
concentration is low (Oh and Cavendish, 1983). Psyllos and Philippopoulos (1993)
numerically studied the performance of monoliths with various parabolic axial
catalyst distributions, and showed that certain distributions have shorter warm-up
periods than the uniform one. Tronci et al. (1999) demonstrated theoretically that
further improvements can be obtained by optimizing the axial catalyst distribu-
tion. By using the optimal distribution, lightoff time was decreased significantly as
compared to the uniform distribution. A two-zone converter was also investigated,
and resulted in similar performance to that of the optimal distribution.
A two-dimensional catalyst distribution pattern, both axially along and radially
across the monolith, was investigated by Naoki et al. (1996), and is shown in Figure
6.5. The quantity of noble metal was six times the base loading in the outer region
of the first zone, equal to the base loading in the inner region of the first zone,
and one-half the base loading in the second zone. This catalyst distribution was
shown theoretically to exhibit higher conversion efficiency during cold start than
did a uniform monolith with the same total catalyst amount. The investigators
considered radial variations in inlet velocity, inlet temperature, radial heat transfer,
and diabatic operation. High loading in the first zone was found to improve cold-
start behavior, in agreement with the discussion above. Concentrating the catalyst
towards the central region of the first zone improved conversion at the early stage
of warm-up, while concentrating it towards the periphery, as shown in Figure 6.5,
increased conversion at later stages of the warm-up, after lightoff of the central
region. The latter design exhibited the best performance and could be improved
further by insulating the monolith.
6.2 Hydrotreating Catalysts

Hydrotreating catalysts are used in the petroleum industry to remove various
compounds (e.g. metals, sulfur) from the bottom fraction of distilled crudes. This
Figure 6.6. Schematic representation of pore-mouth

plugging of catalyst during hydrodemetallation.
(From Limbach and Wei, 1988.)
is required in order to prevent these contaminants from poisoning the catalysts

used in subsequent processes. In hydrodemetallation, the metals are removed
by irreversible adsorption on the catalyst. In contrast to previous situations, this
deactivating process is actually desired, since the goal is to maximize the amount
of metals deposited on the catalyst.
Due to strong diffusional limitations, metal deposition primarily occurs at the
pore mouth, which hinders the entrance of other reactant molecules, thus leaving
a substantial fraction of the catalyst in the internal part of the pellet unutilized.
This process is illustrated schematically in Figure 6.6, where it is shown that when
the thickness of the deposit equals the difference between the pore radius Rp and
the reactant molecule radius Rm , then the transport through the pore mouth is
inhibited and the pellet becomes inactive. For first-order deposition kinetics, the
time t when such pore-blocking occurs can be estimated from the relationship
t = ρ d (Rp − Rm )/kC (6.1)
where ρd is the deposited layer density, C is the reactant concentration (kg/kgoil ),

and k is the rate constant of the deposition process based on catalyst surface
(kgoil /m2 ·s).
In order to use the catalyst more effectively, various strategies have been ex-
plored, involving optimization of pore size distribution and of catalyst shape. In a
detailed theoretical study, Limbach and Wei (1988) investigated the performance
of catalyst pellets with nonuniform activity distributions, using a diffusion–reaction
model. Hydrodemetallation was simulated by a single isothermal first-order reac-
tion. It was assumed that the intrinsic reaction rate per unit active surface area is
not affected by deposition. The change of active surface area as well as the decrease
of the effective diffusion coefficient with deposition was accounted for, because,
as illustrated in Figure 6.6, the deposited metal decreases the cross-section area
of the pores.
In the uniform catalyst pellet, the pinch point (where pore plugging occurs)
is located at the pore mouth. In this case, the interior portion of the catalyst is
not properly utilized. Therefore, in order to maximize the total metal deposit,
the pellet must fill from the center outwards. This can be accomplished using an
initial catalyst distribution increasing towards the pellet center, which shifts the
pinch point from the pellet surface towards the center. This catalyst distribution
also compensates for low reactant concentration in the pellet interior (the Thiele
modulus of hydrodemetallation is typically large), and increases the amount of
Figure 6.7. Metal deposition profiles resulting from optimal

and uniform activity profiles at various values of dimension-
less operating time . Dimensionless time is based on useful
catalyst lifetime. (From Limbach and Wei, 1988.)
deposition. No restrictions were placed on either the local or the overall catalyst
loadings in seeking an optimal solution that maximizes the amount of contam-
inant metal deposition over the useful catalyst life. The optimal catalyst distri-
bution was calculated numerically. The performance of a nonuniform catalyst
was compared with that of a uniform catalyst which was already optimized for
catalyst loading and pore diameter. As expected, the obtained optimal distribu-
tion exhibited a sharp maximum towards the center of the pellet followed by
a decrease towards the external surface. This distribution provided a substan-
tial increase in the overall amount of metals deposited, as shown in Figure 6.7.
As approximations of the optimal distribution, edge-reduced (where the catalyst
distribution was uniform, but reduced only at the outer edge of the pellet) and
egg-yolk catalysts were also considered. It was found that the total metal deposited
in a uniform pellet could be increased by nearly 25% utilizing the edge-reduced
profile.
Chiang and Tiou (1992) extended the above ideas to the fixed-bed reactor. In
order to explore the effects of nonuniform catalyst distribution, they considered
three specific cases: linear (increasing towards the center), edge-reduced, and egg-
yolk catalysts, with the same total loading. The evaluation was based on the reactor
metal deposition capacity at the point where the metal compound concentration in
the reactor effluent reached 30% of the inlet value. The linear profile exhibited the
largest improvement in metal deposit (almost 50%) over the uniform distribution.
The edge-reduced and egg-yolk catalysts also showed significant improvements
over the uniform pellets. For each catalyst distribution, the initial pore size of the
pellet was optimized. The optimal pore sizes of the nonuniform distributions were
smaller than those of the uniform one, because they increased the pore surface area
for metal deposition while simultaneously delaying the plugging process. Further
improvement was realized with two- and three-zone reactors, but the additional
improvement was smaller for the nonuniform distributions.
The case of a moving-bed reactor was also investigated (Chiang and Fang,
1994), using two families of active catalyst distributions: linear and two-step. Pore
size and active catalyst distribution were optimized simultaneously. The benefit
of using nonuniform distributions over the uniform was small, in contrast to the
fixed bed reactors discussed above, and can be attributed to the short residence
time of catalyst in the moving-bed reactor. The optimal pore size and total metal
deposit were both larger for the fixed bed than for the moving bed.
In all the above studies pore connectivity was not taken into account. Arbabi
and Sahimi (1991a,b) developed a percolation model which predicts the occurrence
of a transition point (the percolation threshold) where the previously connected
cluster of unplugged pores becomes disconnected, so that macroscopic transport
and reaction are no longer possible. Similar catalyst performance improvements
obtained by Limbach and Wei (1988) for nonuniform catalyst distribution were
also observed. Nonuniform catalyst distributions with activity increasing towards
the pellet center lead to increased metal deposition capacities.
For hydrodesulfurization (HDS), Asua and Delmon (1987) investigated the-
oretically the behavior of catalysts containing molybdenum and cobalt sulfides.
In this case, catalyst performance optimization can be pursued by varying both
the overall catalyst concentration (i.e. Mo+Co) as well as its composition (i.e.
Mo/Co). They used a detailed kinetic model which accounted explicitly for the
catalyst composition, as well as for the concentration of active sites, and showed
that the catalyst performance could indeed be improved by using nonuniformly
distributed catalyst pellets.
Goula et al. (1992a) studied experimentally the hydrodesulfurization of thio-
phene on eggshell, egg-white, egg-yolk, and uniform MoO3 /Al2 O3 catalysts. All
catalysts had similar loadings (12–14% MoO3 ) and dispersions (0.098–0.134), but
different numbers of active sites. The highest activity for HDS was found for
the egg-white catalysts, and it was related to the number and type of active sites
rather than the Mo distribution. However, from powder experiments it was es-
tablished that diffusional resistances were present, and the effectiveness factor
followed the trend η eggshell > η egg-white > η egg-yolk . On the other hand, the selec-
tivity towards butane, which is formed by hydrogenation of the unsaturated hy-
drocarbons produced by HDS, was affected by the catalyst distribution and fol-
lowed the trend Segg-yolk > Segg-white > Seggshell . This behavior was related to the
residence time of unsaturated hydrocarbons inside the pellets. In addition, the
uniform catalyst exhibited the lowest activities for both hydrodesulfurization and
hydrogenation, and this was attributed to partial clogging of the pores during
preparation.
6.3 Composite Zeolite Catalysts 119
6.3 Composite Zeolite Catalysts

Composite zeolite catalysts consist of zeolite catalyst particles embedded in a sup-
port (e.g. silica–alumina), which may exhibit some modest catalytic activity itself.
The support has much larger pores, and therefore diffusion is faster in the sup-
port than in the active catalyst. This is the main advantage of composite catalysts,
since the reactants have easier access to the active catalyst particles than in the
case where the entire pellet is composed of the active catalyst alone (Ruckenstein,
1970; Smirniotis and Ruckenstein, 1993). In the presence of pore-mouth poisoning,
these systems exhibit a further advantage, which has been discussed by Varghese
and Wolf (1980) with reference to uniformly distributed pellets. It was shown that
by increasing the poison diffusivities and poison-getter capacity of the support,
the catalyst lifetime can be increased relative to pellets which contain the active
catalyst alone.
Dadyburjor (1982) investigated the possibility of improving composite pellet
activity through the use of nonuniform catalyst distribution. In contrast to the
above studies, the matrix was considered to exhibit some catalytic activity. How-
ever, it was lower than that of the embedded particles, as is the case for hydro-
carbon catalytic cracking on zeolite particles embedded in silica–alumina. The
diffusivity in the matrix was larger than in the catalyst particles, and a first-order
reaction was considered. The simulations indicated that the overall reaction rate
was in the order uniform > concentrated egg-yolk > concentrated eggshell, where
“concentrated” indicates that the catalyst layer is undiluted zeolite. This behavior
is a consequence of the matrix reactivity. The matrix in fact contributes significantly
to the overall reaction rate for the uniform and egg-yolk but not for the eggshell
catalyst, because in the last significant diffusional resistances are introduced by
the external zeolite layer.
The effect of a uniform deactivation process, mainly due to coking, was also
investigated using the “snapshot” approach, i.e. identifying the optimal zeolite
distribution at various deactivation levels, which were simulated by using pro-
gressively smaller zeolite reactivity and matrix diffusivity. For a fresh catalyst,
the optimal distribution was monotonically decreasing from the pellet center
to zero at the pellet surface. When diffusional resistances were increased,
either by increasing the pellet size or by simulating the effect of coking, the
optimal distribution changed shape and exhibited a maximum at a subsurface
location. For a first-order isothermal reaction, the eggshell catalyst would give
maximum conversion if the zeolite and matrix diffusivities were the same. How-
ever, for this system there is an interplay between two levels of diffusion resis-
tances and reactivities. At smaller pellet sizes, the higher reactivity of zeolites
overshadows its lower diffusion coefficient. At larger pellet sizes, it is more im-
portant for the matrix to allow the reactant to diffuse inside the pellet than for
the zeolite to convert as much as possible of the reactant at the pellet surface.
However, at least for the explored parameter values, the optimal distribution
resulted only in modest improvement in overall reaction rate over the uniform
pellet.
The improvement in selectivity through the use of nonuniform active catalyst

distribution was larger, and ranged from 10% to 100% relative to the uniform dis-
tribution for the range of parameters investigated (Dadyburjor, 1985). Parallel and
series reaction systems with first-order kinetics were examined to represent hydro-
gen transfer, β-scission, and coke formation, which typically occur during catalytic
cracking. The performance of concentrated eggshell, egg-white, and egg-yolk and
of uniform and diluted egg-white distributions was examined. In the last case, the
internal layer was formed by a mixture of zeolite and matrix (silica–alumina). In all
distributions, the total amount of zeolite was constant. Optimization was carried
out within the family of diluted egg-white catalysts by computing the values of
inner and outer radii of the internal layer leading to maximum selectivity or yield.
Note that by properly selecting the two radius values, the diluted egg-white dis-
tribution reduces to each of the other four distributions mentioned above. In the
parallel reaction system where the desired product is formed by hydrogen transfer
and the undesired product by β-scission, the concentrated eggshell distribution
offered the highest selectivity, while diluted eggshell distributions optimized yield.
In the series reaction scheme, A→C→D, where the desired product C is formed
by hydrogen transfer and further reaction results in undesired coke precursors D,
selectivity was maximized by diluted eggshell distributions. For cases where a di-
luted eggshell distribution was optimal, the thickness of the diluted layer became
smaller with deactivation.
Lee and Ruckenstein (1986) represented the catalytic cracking process using
the reaction scheme
A→B→C (6.2)
→
→
where component A is oil, C is gasoline (which is the desired product), D is gases

which result from the secondary cracking of gasoline, and B is an intermediate
compound. If formation of the intermediate B is neglected, the above network
simplifies to a series reaction scheme. In this case the rate constant for A→C
was taken larger in the zeolite than in silica–alumina, while the rate constant
for C→D was taken smaller in the zeolite than in silica–alumina. It was found
that a diluted eggshell distribution of the zeolite was preferable to a uniform
or egg-yolk distribution for yield and selectivity maximization. This leads to a
high rate of formation of C at the zeolite location, and a significant part of C
can diffuse out of the pellet without being transformed to D in the core region.
As mentioned above, a similar diluted eggshell distribution was found to be opti-
mum for selectivity maximization in a consecutive reaction scheme by Dadyburjor
(1985), even though both rate constants were larger in the zeolite than in the ma-
trix. This indicates that if the rate constant for the desired product is greater in
the zeolite than in the matrix, the optimal distribution type is the same regard-
less of the rate constant of the secondary reaction being larger or smaller in the
zeolite.
When the intermediate B is included in the reaction set, then for the case where
the rate constants of all reactions but B→C are greater in the silica–alumina
matrix than in the zeolite, the diluted egg-yolk distribution leads to higher yield
and selectivity than the uniform distribution. This occurs because the formation
of component B in the outer region can lead to a greater amount of desired C in
the inner region. The optimum radius of the diluted core and the optimum yield
both depend on the values of the parameters. When the overall volume fraction of
silica–alumina was also considered as an optimization variable, it was found that
yield exhibited a maximum with respect to both the diluted-core radius and the
volume fraction of silica–alumina.
Dadyburjor and coworkers (Martin et al., 1987; White and Dadyburjor, 1989;
Dadyburjor and White, 1990) studied extensions of the above schemes including
changes in reaction orders as well as position-dependent rate constants and diffu-
sivities arising from nonuniform poisoning. Since these variations induce complex
interactions between transport and reaction in the zeolite and matrix, a variety
of zeolite distributions were found to be optimal, depending on the specific case
investigated.
6.4 Immobilized Biocatalysts

After immobilized enzymes were successfully industrialized by the Tanabe
Seiyaku Company in 1969 (Tosa et al., 1969), utilization of biocatalysts such as
enzymes, microorganisms, and cells increased sharply. Many papers have since
been published regarding immobilization methods and applications of immobi-
lized biocatalysts, and various processes have been developed at an industrial scale
(cf. Furusaki, 1988; Chibata et al., 1992; Lilly 1996). Mass transfer resistance in
immobilized enzymes can be significant because the reactions take place in liquid
(usually aqueous) phase and reactant transport inside the support is generally
slower than reaction. The importance of such resistance for catalyst and reactor
performance has been widely recognized in the biochemical engineering literature
(cf. Vieth et al., 1973; Radovich, 1985; Buchholz and Klein, 1987; Furusaki, 1988).
The most common kinetic expression to describe a biochemical reaction is the
Michaelis–Menten kinetics,
rmax C
r= (6.3)
Km + C
where rmax is the maximum reaction rate and Km the Michaelis–Menten constant.
This kinetics is monotone increasing in substrate concentration, behaving as first-
order at low and as zero-order at high concentrations. Thus, for isothermal condi-
tions, mass transfer resistance can only be detrimental. Based on the theoretical
results discussed in Chapter 2, we expect the optimal distribution to be a surface
Dirac-delta, and its advantage over a uniform distribution is more significant at
low substrate concentration and large Thiele modulus. Accordingly, most studies
which consider nonuniform catalysts have focused on eggshell distributions.
One of the early studies was by Horvath and Engasser (1973), who investigated
theoretically the properties of pellicular heterogeneous catalysts, which consist of
a fluid-impervious core supporting a spherical annulus of the catalytically active
porous medium. This configuration is equivalent to an eggshell enzyme distribu-
tion. For Michaelis–Menten kinetics under isothermal conditions, it was shown
that a fixed amount of enzyme gives higher overall reaction rate when it is con-
fined to the outer shell than when it is distributed uniformly in a porous pellet of
the same diameter. As expected, significant improvements in overall reaction rate
are realized by the eggshell catalyst for high Thiele modulus and sufficiently low
reactant concentration. The latter occurs at the end of batch operation, when high
reactant conversions are required. These conclusions were confirmed experimen-
tally for eggshell and uniformly distributed agarose–micrococcal nuclease in the
hydrolysis of a phosphate compound, which follows Michaelis–Menten kinetics
(Guisan et al., 1987). Under conditions of low reactant concentration and high
Thiele modulus, the highest increase of effectiveness factor obtained theoretically
was 35%, while a 16% improvement was observed experimentally.
Higher effectiveness factors for eggshell immobilized trypsin on derivatized
glass carriers, as compared to a uniform distribution, were demonstrated experi-
mentally and theoretically by Buchholz (1979) and Borchert and Buchholz (1979,
1984). For the hydrolysis of low concentration N-α-benzoyl-L-arginine ethyl ester
solutions, experimental effectiveness factors for eggshell distributions were higher
by up to 50%; when the higher-molecular-weight reactant casein was used, the
effectiveness factor was higher by up to 100%, and this was attributed to large in-
traparticle concentration gradients. The improvements in effectiveness were more
significant when larger support particles were used, since the Thiele modulus was
then higher.
Because of the sensitivity of enzyme activity to pH, another reason for infe-
rior performance of the uniform distribution could be slow proton diffusion from
inner core of the particle to solution. Carleysmith et al. (1980a) showed that for
deacylation of benzylpenicillin catalyzed by penicillin acylase, slow proton diffu-
sion resulted in low reaction rates even when reactant diffusional limitations were
decreased by the use of small particles. It was postulated that a pH gradient ex-
isted in the beads even in the presence of a buffer. On the other hand, for eggshell
distributions, all the enzyme probably operated near its optimum pH.
Subsurface enzyme distributions can confer a measure of protection against
degradation by local concentrations of harsh reagents which may need to be added
during reaction. Carleysmith et al. (1980b) showed experimentally for the hydrol-
ysis of benzylpenicillin that when the enzyme (penicillin acylase) was immobilized
in a subsurface layer, the particles retained their initial activity to a larger extent
after three consecutive batch reactions than with a shell-type enzyme distribution.
This was attributed to the fact that the enzyme was protected from localized re-
gions of high pH that resulted when alkali was added to maintain the pH of the
reaction medium constant.
Park et al. (1981) extended Horvath and Engasser’s work by studying theo-
retically the behavior of eggshell, egg-yolk, and uniform enzyme distributions for
Michaelis–Menten kinetics with substrate inhibition

rmax C
r= (6.4)
Km + C + Kis C 2
and product inhibition
rmax C
r= . (6.5)
Km [1 + (Cproduct /Kip )] + C
For product-inhibited reactions the eggshell distribution gave a higher effective-

ness factor, similarly to the kinetics (6.3) – as is expected, since both rates show
the same monotonic dependence on reactant concentration. In addition, the ki-
netics (6.5) exhibits a negative-order dependence on product concentration. This
also favors eggshell catalysts, because the product can diffuse easily to the bulk
fluid and hence its concentration in the enzyme layer is kept small. The substrate-
inhibited kinetics (6.4) exhibits a reaction order with respect to substrate which
varies from +1 at low to −1 at large concentrations. Thus, for certain conditions,
the best catalyst performance is given by egg-white or egg-yolk distributions, since
the diffusion barrier of the outer inert section reduces the inhibition effect. Re-
actions with similar kinetics were studied theoretically by Juang and Weng (1984)
in pellets with increasing, decreasing (towards the center), and uniform enzyme
distributions. As expected, the decreasing profile gave the highest effectiveness
factor for the kinetics (6.3), while the increasing one was more effective for the
substrate-inhibited kinetics in certain ranges of Thiele-modulus values. Follow-
ing similar arguments to those in section 2.1.1, it was shown (Morbidelli et al.,
1984b) that the optimal enzyme distribution is given by a Dirac-delta. For the
kinetics (6.3) it is always located at the particle external surface, while for the
substrate-inhibited kinetics (6.4) it may be located at any position within the sup-
port, depending on the reaction conditions. This finding is in agreement with the
results discussed above for different types of nonuniform distributions.
Cases where enzyme deactivation was present were investigated for the kinet-
ics (6.3) by Juang and Weng (1984) and Hossain and Do (1987). The decay of
the enzyme activity also depended on substrate concentration, following a similar
kinetic expression. When operating in the diffusion-limited regime, distributions
with decreasing (towards the center) activity showed the best performance at
short operation times, but for longer periods, higher effectiveness factors were
obtained with uniform or increasing enzyme distributions. This occurs because as
time elapses, the immobilized enzyme near the surface deactivates due to high
substrate concentration, thus allowing more substrate to diffuse into the interior
of the particle. Consequently, enzyme in the interior of the particle, which was
previously unutilized because of lack of substrate, gradually plays a role in the re-
action process. According to the analysis of selective deactivation systems (section
4.2), the optimal enzyme distribution should be a Dirac-delta. For short operating
times and the kinetics (6.3), a surface Dirac-delta should provide the best over-
all performance, and this is consistent with the above findings (Juang and Weng,
1984; Hossain and Do, 1987). The fact that uniform and increasing distributions
show higher conversions than decreasing distributions at large operating times in-
dicates that the optimal enzyme concentration is an appropriate subsurface Dirac
distribution.
Chung and Chang (1986) compared theoretically the performance of nonuni-
form biocatalysts for first-order kinetics in the presence of two different deactiva-
tion mechanisms representing the effects of temperature and time. The first was
based on a reversible thermal denaturation model, according to which enzyme
activity was an exponentially decreasing function of temperature (Ollis, 1972).
This, combined with the usual Arrhenius dependence, gives rise to a maximum
in the first-order reaction rate constant as a function of temperature. For the sec-
ond deactivation mechanism, the reaction rate constant was assumed to decay
exponentially with time. Three distributions with the same amount of enzyme
were considered: uniform, eggshell, and egg-yolk. Egg-yolk immobilized enzymes
had the smallest effectiveness factors, but they showed the best stability under
temperature fluctuations. The egg-yolk distribution also exhibited better stability
than eggshell catalysts over time, because in the former the effect of rate-constant
decrease is counterbalanced by better utilization of the enzyme arising from the
concomitant decrease of the Thiele modulus. Thus, stability with time increased
in the order eggshell to uniform to egg-yolk.
6.5 Functionalized Polymer Resins

Improvement of catalyst performance through an appropriate nonuniform distri-
bution of active element has also been investigated in the case of cation-exchanged
polymer resins. They were originally developed for ion-exchange processes, such
as those involved in water treatment or hydrometallurgy. These resins consist of
a crosslinked polymer matrix, which is functionalized by acid groups covalently
bonded to the polymer. The most common are made of a styrene–divinylbenzene
copolymer, whose degree of crosslinking increases with divinylbenzene amount
(usually 2–20 wt%). Functionalization is achieved by attaching sulfonic groups to
the pendant phenyl groups. Due to the acid characteristics of the former, these
ion-exchange resins can be used as supported acid catalysts for a variety of chem-
ical reactions, including esterification, etherification, dehydration, alkylation, and
isomerization (cf. Sherrington and Hodge, 1988).
As compared to the common homogeneous catalytic processes involving min-
eral acids such as sulfuric and hydrochloric acids, functionalized resins offer the
advantages of heterogeneous catalysts: elimination of waste disposal problems,
and catalyst reuse. However, it is worth noting that the activities and selectivities
of the two catalytic systems are in principle different and depend on the specific
reacting system and operating conditions.
6.5.1 Preparation of Nonuniformly Functionalized Resin Particles

Klein et al. (1984) showed that it is possible to prepare polymer particles with
nonuniform distribution of sulfonic groups. The form of distribution depends on
resin type (gelular or macroreticular) and sulfonation conditions. Gelular resins,

which are in the form of compact beads of crosslinked polymer with no permanent
porosity, can be penetrated by low-molecular-weight species only by swelling. This
is a selective process, where diffusion and adsorption of species with higher affinity
to the resin are favored. Thus, nonsulfonated resins preferentially sorb organic
species, while sulfonated resins preferentially sorb polar species. This is important
in organic synthesis, since it leads to different compositions of the reacting mixture
in the external fluid phase and inside the resin particle, i.e. in the reaction locus.
Therefore, by controlling the amount and distribution of sulfonic groups in the
resin, one can control the selective sorption process and hence the composition
in the reaction locus, which ultimately affects reaction rate and selectivity. The
above clearly indicates that the performance of functionalized resins may differ
significantly from that of homogeneous catalysts.
Macroreticular resins consist of a large number of gelular microparticles
(∼100 nm in diameter) separated by macropores through which diffusion takes
place. Each of the gelular microparticles can swell, leading to selective sorption
as discussed previously. On the other hand, the diffusion through the macrop-
ores is not selective, since intraparticle and external fluid phases are similar in
composition. In the presence of chemical reaction, competition between diffu-
sion and reaction arises, which can lead to concentration gradients of the reacting
species in the resin particle, i.e. in the macropores. In addition, localized concen-
tration gradients inside the microparticles develop due to the selective sorption
process.
The swelling and diffusion–reaction phenomena described above take place
during the sulfonation process, which is utilized to prepare different types of
nonuniformly functionalized resin particles. For a gelular particle, surface dis-
tribution of acid sites is obtained by sulfonating the resin without a swelling agent.
When such an agent is used, the intraparticle profile of sulfonic groups is the result
of the relative rates of the sulfonation reaction and of diffusion of the sulfonating
agent driven by the swelling. If the diffusion is slower, decreasing concentration
profiles of the sulfonic groups from surface to center are obtained. In the oppo-
site case, as well as for sufficiently long sulfonation times, a uniform distribution
results.
The situation is more complex for macroreticular resins. As discussed above,
different concentration profiles can develop inside the macroparticle (i.e. in the
macroporous phase, and therefore also on the microparticle surface at the same
location) and the gelular microparticles. In particular, Klein et al. (1984) obtained
four different pairs of profiles in the macro- and microparticles, as shown schemat-
ically in Table 6.1. In the first case (type A), which corresponds to typical commer-
cial resins, the distribution was uniform at both scales. This was obtained by using
a suitable swelling agent and sufficiently long sulfonation times. In the remaining
three cases, swelling agents of increasing strength were used to increase the charac-
teristic diffusion time in the microparticles up to the point where it became compa-
rable to that in the macropores. Thus, for type B particles, where H2 SO4 as sulfonat-
ing agent and CH3 NO2 as swelling agent were used, diffusion in the microparticles
Table 6.1. Nonuniform distribution of sulfonic groups in resin macroparticles

and microparticles (From Klein et al., 1984.)
Distribution
Macroparticle Microparticle Synthesis Type Figure
H2 SO4
completely A
sulfonated
H2 SO4 /CH3 NO2 B 6.8
H2 SO4 /C2 H4 Cl2 C 6.9.a
SO3 /N2 D 6.9.b
was much slower than in the macropores. Therefore a uniform profile was obtained
in the macroparticle, while the microparticles exhibited a surface distribution. For
particles of type C, using C2 H4 Cl2 as swelling agent, decreasing profile towards the
center was obtained at both scales. Finally, for particles of type D, where the most
active sulfonating agent SO3 was used (in a gas-phase fluidized bed with nitrogen)
together with a relatively short sulfonation time, the sulfonic group concentration
profile was steeper in the macropores than in the microparticles.
Verification of the distributions discussed above was obtained by measuring
the concentration of sulfonic groups by energy-dispersive X-ray analysis (EDX)
of cross-sectional areas of resin particles (Bothe, 1982; Klein et al., 1984). The
available resolution permitted measurement of profiles only for the macroparti-
cles, not for the microparticles. In Figure 6.8 sulfonic-group concentration profiles
measured as a function of time during sulfonation of type B particles are shown.
The profile in the macropores is indeed uniform, and concentration increases with
time as expected. Although the microparticle profiles cannot be measured, results
reported by Ahn et al. (1988) under similar conditions show that gelular resins
are not swollen and that sulfonation occurs only at the external surface. Finally,
Figure 6.9 confirms the sulfonic groups’ concentration profiles indicated earlier
for type C and D particles, respectively (see Table 6.1).
6.5.2 Applications to Reacting Systems

The effects of nonuniform distribution of sulfonic groups on the performance of
resin particles have been demonstrated for several systems. Klein et al. (1984)
Figure 6.8. Scanning electron micrographs and EDX

line scans of the cross section of resin particles sulfonated
with H2 SO4 /CH3 NO2 after 2, 5, 10, and 22 h. (From Klein
et al., 1984.)
Figure 6.9. Scanning electron micrographs and EDX

line scans of the cross section of a resin particle sul-
fonated with (a) H2 SO4 /C2 H4 Cl2 and (b) SO3 in a bed
fluidized with N2 . (From Klein et al., 1984.)
used the alkylation of benzene with propylene as a model reaction. Due to the
nonpolar nature of the reactants, the gelular microparticles were not swollen, and
therefore the acid groups located inside the microparticles were not accessible
to the reacting liquid. This explains the higher activity observed for the catalyst
of type C in Table 6.1, as compared to uniform particles A, as well as to particles
of type D, in which the largest fraction of acid groups are inside the gelular phase
and are hence inaccessible.
Ahn et al. (1988) considered the isomerization of 1-butene as a model reac-
tion to investigate the performance of three types of macroporous resin particles.
Two of these, obtained by sulfonation of preswollen resin particles, were uniform
but with different total loading. The third was of type B in Table 6.1, i.e., it con-
tained sulfonic groups only on the surface of the microparticles, but uniformly
distributed throughout the entire macroparticle. The results obtained were inter-
preted through a two-phase model, where the simultaneous diffusion and reac-
tion processes were described at both the microparticle and macroparticle levels.
In particular, the overall effectiveness factor of the macroporous particle, η was
computed from
η = ηα [β + (1 − β)ηι ] (6.6)
where ηα and ηι are effectiveness factors accounting for diffusion in the macrop-
ores and the microparticle, respectively, while β is the fraction of sulfonic groups
on the external surface of the microparticles, which are accessible immediately
through diffusion in the macropores. The experiments were found to be consis-
tent with the model results, leading to values of the microparticle effectiveness
factor in the range 0.1–0.3 and to a reaction order with respect to sulfonic-group
concentration equal to 2.4. The latter was explained by considering that the rate-
determining step of the isomerization involved 2–3 sulfonic groups, in agreement
with the results reported by Gates et al. (1972) for t-butanol dehydration. Catalyst
B showed the best performance, since for a given loading of acid groups the local
concentrations were higher and the reaction was faster.
Chee and Ihm (1986) investigated the effect of nonuniform distribution of ac-
tive sites on catalyst deactivation during the gas-phase ethanol dehydration. In
this system the produced water molecules interact with pairs of sulfonic groups,
either forming hydrogen bonds or reacting with them to cause resin desulfonation.
Both mechanisms were considered in a deactivation kinetic expression which was
second-order with respect to sulfonic-group concentration. The two-phase model
mentioned above was used to simulate the deactivation process observed exper-
imentally in two types of sulfonated membranes. The macroporous membranes
utilized had the same total loading of sulfonic groups but different distributions:
one was uniform, while in the other only the external surface of the microparticles
was functionalized, i.e. type A and B in Table 6.1, respectively. The observed de-
activation patterns were quite different, and the nonuniform membrane exhibited
the fastest deactivation rate due to the high surface concentration of acid groups.
The experimental observations were in qualitative agreement with the two-phase
model.
Nonuniform sulfonic-group distributions can also be employed to improve the

selectivity in a multiple reaction system, by favoring the desired reaction relative
to the undesired ones. Widdecke et al. (1986) studied the ether cleavage reaction
C4 H9 OCH3 → C4 H8 + CH3 OH (6.7)
which produces isobutene and methanol. Undesired reactions are mainly metha-
nol dehydration and isobutene dimerization,
2CH3 OH → CH3 OCH3 + H2 O (6.8)

2C4 H8 → C8 H16 (6.9)
which are both catalyzed by acids. When a uniform resin is employed, the nonpolar
ether reactant does not enter the sulfonated gelular microparticles, and the main
reaction is restricted to the surface active groups in the macropores. However, the
consecutive reactions, and particularly reaction (6.8) involving a polar reactant,
take place also in the microparticles. Accordingly, when using a resin sulfonated
only on the external surface of the microparticles, i.e. type B in Table 6.1, the
product of the main reaction diffuses out of the macropores, reaction (6.8) is prac-
tically eliminated, and nearly 100% selectivity to the desired product is achieved
(Widdecke et al., 1986).
Similar selectivity issues are pertinent to esterification and transesterification re-
actions, where an ester is produced by reacting an alcohol with an acid or another
ester, respectively. These are large families of reactions, relevant to various prod-
ucts of industrial interest (Lundquist, 1995). In order to achieve satisfactory reac-
tion rates, high temperatures (60–120◦ C) must be employed, which unfortunately
give rise to undesired reactions catalyzed by the sulfonic groups. As mentioned
above, these include the production of undesired ethers by alcohol condensation.
Additionally, in the case of secondary or tertiary alcohols, dehydration reactions
may also occur, leading to olefinic by-products. When uniform resins are utilized,
the polar alcohol molecules have a strong tendency to enter the sulfonated gelular
microparticles, where both the above undesired reactions can take place. Hence,
similarly to ether cleavage, selectivity is improved by confining the sulfonic groups
to the external surface of the microparticles. There they are equally accessible by
the alcohol as well as the other reactant, i.e. acid or ester for esterification and
transesterification, respectively.
Gelosa et al. (1998) studied the esterification of phthalic anhydride with ethanol,
using type A and B resins (see Table 6.1). The reaction system consists of the two
consecutive reactions
phthalic anhydride + ethanol → monoethylphthalate

monoethylphthalate + ethanol %& diethylphthalate + water.
The first reaction is rapid and noncatalytic, while the second requires acidic cat-
alyst. In Figure 6.10, conversion as a function of time for three different catalytic
systems is shown. Curve 1 corresponds to homogeneous catalysis, obtained using
a 95% solution of sulfuric acid, while curves 2 and 3 correspond to two different
Figure 6.10. Conversion of monoethylphthalate as a

function of time for three different catalytic systems:
(1) sulfuric acid, (2) type A resin, (3) type B resin.
(From Gelosa et al., 1998.)
types of sulfonated resin particles, i.e. uniform (type A in Table 6.1) and surface
distribution (type B), respectively. The conversion profiles are similar as shown
in Figure 6.10, but show significant differences in selectivity. In particular, for the
homogeneous catalyst no by-products are formed, since the alcohol is contacted
by the acid catalyst only in the presence of the monoethylphthalate, in which
case only the esterification reaction is favored. However, when the uniform resin
is utilized, ethanol accumulates in the gelular sulfonated microparticles, where
the monoethylphthalate is almost completely excluded. This favors the ethanol
condensation reaction, which leads to ∼2.5% ether formation at the end of the
reaction. This reaction is minimized with the surface-sulfonated resins, which lead
to only 0.4% ether formation.
7
Preparation of Pellets with Nonuniform
Distribution of Catalyst
N
onuniform catalyst distributions in porous supports are obtained primarily by
multicomponent impregnation techniques. In general, an intermediate level
of interaction between catalyst precursor and support is required, so that the
precursor can attach to the support, but can also desorb if another competing
adsorbing species is present. Depending upon the interplay between competitive
adsorption and diffusion of the various species in the porous support, a variety
of nonuniform catalyst distributions can be obtained. The above physicochemical
processes are also encountered in chromatographic separations (Ruthven, 1984).
This chapter is divided in two parts. The first deals with adsorption on powders,
while the second is focused on simultaneous diffusion and adsorption phenomena.
Although diffusion–adsorption methods are dominant for the preparation of
nonuniform catalyst pellets, other procedures have also been employed. One such
technique is deposition precipitation in preshaped carriers (De Jong, 1991). It in-
volves deposition inside pellets of insoluble compounds, such as hydroxides which
are formed by a precipitation reaction. The latter can be induced by a change of
solution pH. Immediately after imbibition, a pH profile develops inside the pellets,
which depends on the initial solution pH and the isoelectric point of the carrier.
Since precipitation reactions depend on pH, the insoluble compound distribution
reflects the pH gradient. Hence, by appropriate choice of the impregnation con-
ditions, precipitation can occur in either the inner (egg-yolk distribution) or the
outer (eggshell distribution) region of the pellet. However, preparation of eggshell
catalysts leads to the problem of precipitation outside the pellets. Therefore this
method is especially suited for preparation of egg-yolk distributions, and has been
utilized for preparation of Mo/SiO2 , Cu/γ -Al2 O3 , and Ag/γ -Al2 O3 catalysts.
In other preparation techniques, the nonuniform catalyst pellet is obtained by
assembling catalytic and noncatalytic layers. One such method is based on cata-
lyst powder granulation. This can be realized by a two-stage fluidized/spouted bed
(Scheuch et al., 1996). Catalyst powder premixed with a binder is deposited grad-
ually on moist seeds (typically 0.5–2-mm diameter) of catalyst support, ultimately
providing layered spherical particles. This method can in principle be extended for
preparation of egg-white catalysts by using the corresponding eggshell catalysts
131
132 Preparation of Pellets with Nonuniform Distribution of Catalyst
as seed granules and depositing inert powder. Similarly, egg-yolk catalysts could
be produced using uniform spherical catalyst particles as seed granules.
Coating is another technique that has been used for the preparation of egg-
shell catalysts (Pernicone and Traina, 1984; Stiles and Koch, 1995). The coating
equipment generally resembles pill-coating devices used in the pharmaceutical
industry. The cores are coated in a rotating drum, with an appropriate catalytic
paint. The thickness of the layer can be controlled by the amount of paint slurry
introduced in the drum.
Tableting can also in principle be used to prepare egg-shell catalysts or other
nonuniform distributions. In fact, specially designed presses are used in the phar-
maceutical industry to prepare multilayer tablets. Examples of nonuniform pellet
preparation by pressing together different layers are reported by Wu et al. (1990b)
and Gavriilidis and Varma (1992).
Finally, extrusion can also be adapted for the preparation of nonuniform extru-
dates by appropriate design of the die.
7.1 Adsorption on Powders

Adsorption of metal complexes on oxide powders takes place without any dif-
fusional limitation. The analysis can be based on phenomena which occur at the
interface between the support and the impregnating solution, and has been devel-
oped largely in the colloid and interface science literature. Adsorption isotherm
and surface ionization models have been used in order to quantitatively describe
adsorption of catalyst precursors on oxide surfaces from aqueous solutions, with
different outcomes depending on system and model complexity. The surface ion-
ization models usually contain several parameters which can be adjusted to fit
the experiments. As a consequence, different models can represent experimental
data, with different values of the corresponding parameters. Adsorption isotherm
models usually describe experiments within a limited range of conditions, but
are popular because they are simpler to implement and require fewer parameter
estimations.
In the following we first introduce adsorption isotherm models. Next, we discuss
the effect of impregnation variables on adsorption, with reference to various exper-
imental studies reported in the literature. Finally, we introduce surface ionization
models, which are based on a detailed description of the solid–liquid interface.
7.1.1 Adsorption Isotherm Models

The most common relationship used to describe adsorption of catalyst precursors
at constant pH is the Langmuir isotherm. In this case, the net rate of adsorption
is given by
dn
= k+ C(ns − n) − k− n (7.1)
dt
where n and C are the surface and fluid-phase precursor concentrations respec-
tively, and ns is the saturation capacity. At equilibrium, the amount adsorbed can
be calculated by setting the transient term equal to zero, thus giving the Langmuir
adsorption isotherm
ne KCe
= (7.2)
ns 1 + KCe
where ne is the adsorbed amount, and Ce is the solution concentration, both at
equilibrium. K = k+ /k− is the equilibrium adsorption constant, and a large value
indicates strong adsorption. The Langmuir isotherm was first used to describe
adsorption of gases on solids, but was later extended to describe adsorption from
solutions. It assumes that the surface is energetically uniform, the energy of ad-
sorption is independent of surface coverage, and the solution is dilute, so that the
adsorbed species do not interact with each other. This simple equation represents
well a broad spectrum of experimental adsorption data.
Adsorption of many metal precursors on various supports has been studied
under equilibrium and transient conditions in batch systems. The most widely in-
vestigated system is H2 PtCl6 on γ -Al2 O3 . A Langmuir model of adsorption was
generally found to be suitable for fitting the experimental data (Santacesaria et al.,
1977a; Shyr and Ernst, 1980; Castro et al., 1983; Jianguo et al., 1983; Heise and
Schwarz, 1988; Xidong et al., 1988; Subramanian and Schwarz, 1991; Blachou and
Philippopoulos, 1993; Papageorgiou et al., 1996). Figure 7.1 shows experimental
results of H2 PtCl6 adsorption on γ -Al2 O3 under (a) equilibrium and (b) transient
conditions. The surface saturation coverage ns and adsorption equilibrium con-
stant K can be determined from equilibrium measurements, and the adsorption
rate constant k+ from transient measurements. The surface saturation coverage
obtained in Figure 7.1.a is ns = 1.55 µmol/m2 , and the equilibrium adsorption
Figure 7.1. Adsorption of hexachloroplatinic

acid on γ -Al2 O3 powder: (a) equilibrium ad-
sorption isotherm; (b) transient adsorption
behavior (Cb0 = 0.01287 M). The lines are
the fitted Langmuir adsorption models, and
the symbols denote experimental data. (From
Papageorgiou et al., 1996.)
Table 7.1. Values of adsorption parameters of hexachloroplatinic acid on γ -Al2 O3

in water solution.
Area K k+ ns te
(m2 /g) (l/mol) (l/mol · s) (µmol/g) (µmol/m2 ) (h) Reference
177 459 1.21 265 1.50 3–8 Santacesaria et al., 1977a
150 1,330 110 0.73 2 Shyr and Ernst, 1980
170 31,000 140 0.82 6 Castro et al., 1983
245 275 1.12 6 Jianguo et al., 1983a
7,200 1.70 Scelza et al., 1986
190 1,550 151 0.79 0.3 Heise and Schwarz, 1988
195 110 0.18 290 1.48 2 Subramanian and Schwarz, 1991
227 205 0.90 1 Mang et al., 1993
78 606 2.46 121 1.55 1 Papageorgiou et al., 1996
a Solid was η-Al2 O3
constant is 606 l/mol. From Figure 7.1.b, the adsorption rate coefficient k+ was cal-
culated to be 2.46 l/mol · s. Values of these parameters reported in the literature
range for surface saturation coverages from 0.73 to 1.55 µmol/m2 , and for the ad-
sorption equilibrium constant from 110 to 31,000 l/mol (see Table 7.1). Adsorption
characteristics of other systems that have been investigated include compounds
of Pd on alumina (Sivaraj et al., 1991), Pd on silica and alumina (Schwarz et al.,
1992), Pt on carbon (Hanika et al., 1982, 1983; Machek et al., 1983a,b), Rh on
alumina (Hepburn et al., 1991a), Ni on alumina (Komiyama et al., 1980; Huang
et al., 1986; Huang and Schwarz, 1987; Clause et al., 1992), Ni, Ba on alumina –
one- and two-component impregnation – (Melo et al., 1980b), Cr on alumina
(Chen and Anderson, 1973), Mo on alumina (Wang and Hall, 1980), Cr, Cu on
alumina – one- and two-component impregnation – (Chen and Anderson, 1976),
Cr, Mo, W on alumina – one-component impregnation – (Iannibello et al., 1979),
Co, Mo on alumina – one- and two-component impregnation – (Iannibello and
Mitchell, 1979; Cheng and Pereira, 1987; Hanika et al., 1987a,b; Sporka and
Hanika, 1992), and organic acids on alumina (Kummert and Stumm, 1980; Engels
et al., 1987).
The adsorption capacity of a support can be altered by calcination prior to im-
pregnation (Schwarz, 1992). The number of surface sites on alumina which can
be used for binding the metal precursor changes in a nonmonotonic fashion as a
function of calcination temperature. In particular, a minimum in acidity and sur-
face sites has been observed at about 600◦ C by some investigators (Tanabe, 1970;
Sivaraj et al., 1991). Santacesaria et al. (1977b) found that the amount of plat-
inum adsorbed during hexachloroplatinic acid impregnation on alumina samples
calcined at different temperatures followed the same trend as the acidity.
In addition to the Langmuir isotherm, other models such as the Freundlich
isotherm have been used for fitting adsorption data (cf. Benjamin and Leckie,
1981; Haworth, 1990). This isotherm is described by
ne = KCe1/m (7.3)
where K is a parameter related to the heat of adsorption, and m is a constant. Note

that the amount of solute that can be adsorbed according to (7.3) is unlimited.
7.1.2 Effect of Impregnation Variables on Adsorption

There are various parameters that affect adsorption which simple models like the
Langmuir isotherm do not take into account. These include solution pH, nature of
support, surface heterogeneity, ionic strength, precursor speciation, presence of
extraneous ions, and nature of solvent. When one or more of these parameters are
changed, then even for the same precursor–support pair, rather different values
of the Langmuir model parameters are obtained, as shown in Table 7.1 for the
H2 PtCl6 /γ -Al2 O3 system. Let us now review the effect of each of these parameters
through various experimental evidence reported in the literature.
7.1.2.a Solution pH and Nature of Support

The solution pH can strongly affect adsorption in aqueous solutions. Santacesaria
et al. (1977a) associated H2 PtCl6 adsorption with acid attack on the alumina, dis-
solution of aluminum, subsequent formation of aluminum ions, and their final
readsorption on the support after complexation with the hexachloroplatinate an-
ions. Similar dissolution–reprecipitation of Al with Co, Ni, or Zn ions has also been
observed even for nonaggressive pH, i.e. close to the isoelectric point of alumina
(D’Espinose de la Caillerie et al., 1995). The amount of aluminum ions released
into the solution depends on the crystallinity of the support: the solubility of alu-
mina decreases as its crystallinity increases (Xidong et al., 1988; Subramanian
et al., 1992). Various investigators who studied the effect of pH on chloroplatinic
acid adsorption on alumina found that platinum adsorption shows a maximum
at pH = 3–4 (Heise and Schwarz, 1985; Blachou and Philippopoulos 1993; Mang
et al., 1993; Olsbye et al., 1997). Heise and Schwarz attributed the decrease of
adsorbed platinum with decreasing pH to a decrease of adsorption sites due to
alumina dissolution, and Blachou and Philippopoulos to the pH-dependent for-
mation of platinum complexes with different affinity for the alumina surface.
Shah and Regalbuto (1994) argue that the reduction of platinum adsorption
may be attributed entirely to the increased ionic strength of the solution at low pH,
which leads to “double layer compression” or “electric screening”, so that the equi-
librium adsorption constant is effectively decreased. This means that oxide dis-
solution affects adsorption indirectly, through the ionic strength, and not directly
by removing adsorption sites from the surface. In fact, Agashe and Regalbuto
(1997) formulated a Langmuir model where adsorption equilibrium constants
were calculated based on an overall Gibbs free energy comprising terms corre-
sponding to coulombic attraction, repulsive solvation, and adjustable “chemical”
interaction. This model satisfactorily predicted adsorption of metals as a func-
tion of pH, even without the use of the adjustable “chemical” term for certain
systems, indicating that in those cases adsorption was physical (electrostatic) in
nature.
Contescu and Vass (1987) studied the adsorption of chloro- and aminopalla-
dium complexes on alumina and found that at constant pH the adsorption data
followed the Langmuir isotherm. Both the adsorption capacity ns and the adsorp-
tion constant K changed with pH and showed minima at the isoelectric point of
alumina. For molybdate adsorption on alumina, it has been demonstrated that
the adsorption sites are protonated hydroxyls, whose concentration can be de-
creased by increasing the pH (Spanos et al., 1990a,b; Goula et al., 1992b). Huang
(1975) studied the adsorption of ortho-phosphate on hydrous γ -Al2 O3 from dilute
aqueous solution at constant ionic strength. Adsorption followed the Langmuir
isotherm, except for large values of pH (>10). The adsorption constant K was
found to vary with pH, and the adsorption capacity ns showed a maximum at
pH = 4, decreasing rapidly with increasing pH.
In all systems above, a decrease in anion adsorption was observed at pH larger
than a certain value. This can be explained by invoking the simple electrostatic
model of Brunelle (1978). According to this model, when oxide particles are sus-
pended in aqueous solutions, a surface polarization results in net electrical surface
charge. The type and magnitude of this charge are a function of the pH of the so-
lution surrounding the particle and the nature of its surface. In general, in acidic
solutions the surface is positively charged (S–OH+ 2 ), while in basic solutions the
−
particles carry a negative surface charge (S–O ). This is conveniently expressed
by the following equilibrium reactions (Parfitt, 1976):
S–OH+ % +
2 & S–OH + Hs (Ka1 ) (7.4a)
← Decreasing pH
& S–O− + H+
S–OH % s (Ka2 ). (7.4b)
Increasing pH →
In between these two cases, a pH exists at which the net charge of the surface is zero.
This value is characteristic of the oxide, and is called point of zero charge (PZC).
Some authors use the terms point of zero charge (PZC) and isoelectric point (IEP)
interchangeably. Properly, PZC is associated with a zero overall charge of the sur-
face determined by potentiometric titration, and IEP with a zero electrophoretic
mobility determined by microelectrophoresis (Parfitt, 1976). When the PZC and
IEP points coincide (i.e. pHPZC = pHIEP ), the pHPZC is related to the two intrinsic
acidity constants (7.4) by the following equation (Hohl and Stumm, 1976; Schwarz
et al., 1992; Zhukov, 1996):
pKa1 + pKa2
pHPZC = . (7.5)
2
The protonated or deprotonated surface hydroxyl groups [see (7.4)] can be used to
fix the metal ions to the support surface. In acidic solutions the positively charged
surface will preferentially adsorb anions, while in basic solutions the negatively
charged surface will adsorb cations (Brunelle, 1978; D’Aniello, 1981; Foger, 1984;
Spanos et al., 1991). For example, for anionic adsorption the mechanism would be:
+
S–OH + H+ %
& S–OH2 (7.6a)
y(S–OH+
2)+M
−n % +
& (S–OH2 ) y M−n . (7.6b)
In the absence of chemical reactions, the amounts of anion adsorbed can be con-
trolled by adjusting the pH of the impregnating solution. Equivalently, a similar
mechanism can be postulated for cationic adsorption in a basic solution:
S–OH + OH− % & S–O− + H2 O (7.7a)

y(S–O− ) + M+n %
& (S–O− ) y M+n . (7.7b)
Adsorption of ions according to (7.6b) or (7.7b) may be simplistic, but has

been demonstrated for several systems. However, more complicated mechanisms
can be present, where adsorption of various anions produced during precursor
speciation occurs, as for example during molybdena or chromia adsorption on
alumina. In these cases, the metal is present in oxygen-containing anions, which not
only adsorb on S–OH+ 2 but also react with S–OH sites of alumina (Lycourghiotis,
1995). Only amphoteric oxides such as aluminas, titanias, and chromia have the
ability to adsorb both anions and cations. They are characterized by a pHIEP in
the range of 4–9 (Brunelle, 1978). Therefore, for pH < pHIEP they adsorb anions,
while for pH > pHIEP they adsorb only cations. Acidic oxides such as zeolites,
silica–aluminas, and silicas have low pHIEP , and hence they adsorb cations, while
basic oxides such as magnesia and lanthana have large pHIEP , and they adsorb
anions. A comprehensive collection of isoelectric points of oxides and hydroxides
can be found in Parks (1965). It is worth noting that isoelectric points of different
samples of the same oxide may vary markedly. This has been attributed to factors
such as impurities, surface crystallinity, dehydration, and aging.
The isoelectric point of oxide supports is strongly influenced by the presence
of foreign ions on its surface (Jiratova, 1981; Vordonis et al., 1984, 1986a,b, 1990;
Mulcahy et al., 1987; Akratopulu et al., 1988; Mieth et al., 1990). Thus doping
with suitable foreign ions can be used to influence the amount of ion adsorption
that occurs for a given solution pH. The PZC is also affected by temperature.
Akratopulu et al. (1986) found that precise regulation of the γ -Al2 O3 pHPZC can
be achieved by varying the impregnation temperature. In this way the adsorp-
tion capacity of the support can be altered (Spanos et al., 1990b; Lycourghiotis,
1995). Alternatively, the PZC can be changed by calcination prior to impregna-
tion. Sivaraj et al. (1991) found that the pHPZC of alumina increased gradually
with calcination temperature.
The solution pH affects not only the adsorption equilibrium constant but also
the kinetic constants of the Langmuir model. Adsorption of hexachloroplatinic
acid on alumina was studied by Blachou and Philippopoulos (1993) at two different
solution pH values, chosen in such a way that the same amount of platinum was
adsorbed at equilibrium (pH = 2.8 and 4.5). Adsorption kinetics were faster for
the low-pH solution. This is probably related to the increased rate of formation
of S–OH+ 2 at higher solution acidity, according to (7.6a).
We have so far tacitly assumed that each precursor ion adsorbs on one op-
positely charged binding site. However, in some instances the amount adsorbed
is only a fraction of the available sites (Santhanam et al., 1994; Contescu et al.,
1993a, 1995a). Santhanam et al. (1994) attributed this discrepancy to the large size
of ions which adsorb with one or two hydration sheaths intact, whereas Contescu
et al. (1993a, 1995a) attributed it to steric constraints, which force the geometry
of adsorbing complexes to match the arrangement of charged surface sites.
7.1.2.b Surface Heterogeneity

The electrostatic model of Brunelle (1978), according to which adsorption of
cations/anions takes place in solutions with pH higher/lower than the pHPZC ,
agrees with experimental data for many precursor–support systems (Kim
et al., 1989; Bonivardi and Baltanas, 1990; Mulcahy et al., 1990; Vordonis et al.,
1990; Sivaraj et al., 1991; Clause et al., 1992; Karakonstantis et al., 1992; Spanos
et al., 1992; Zaki et al., 1992; Spielbauer et al., 1993). However, certain systems
do not follow this general rule. For example, adsorption of cations on alumina has
been reported for pH < pHPZC (Hohl and Stumm, 1976; Komiyama et al., 1980;
Huang et al., 1986; Chu et al., 1989; Vordonis et al., 1992; Subramanian et al.,
1992). This behavior can be explained by adsorption if one takes into account sur-
face heterogeneity, i.e. the existence of several distinct types of sites with varying
affinities for adsorbate ions (Hiemstra et al., 1989a,b; Contescu et al., 1993a,b,
1995a,b; Abello et al., 1995). Wang and Hall (1982) suggested that the presence
of different crystal planes of alumina, with different local isoelectric points, was
responsible for the surface heterogeneity, manifested by the observed breaks of
the loading curves for molybdate and tungstate ions as a function of pH. Benjamin
and Leckie (1981) invoked surface-site heterogeneity to explain deviations from
Langmuir behavior. They studied adsorption of Cd, Cu, Zn, and Pb on amorphous
iron oxyhydroxide and obtained data for a wide range of adsorption loadings. At
low loadings all types of sites were available in excess, and the metal ion exhibited
Langmuir adsorption. At higher loadings, deviation from the Langmuir behavior
was observed, associated with a decrease in availability of the strongest binding
sites, leading to a decrease in the apparent adsorption equilibrium constant.
Hiemstra et al. (1989a,b) developed a multisite model for proton adsorption,
assigning different proton affinity constants to different types of surface groups ex-
isting at the solid–solution interface. The sites that develop after proton adsorption
or desorption were considered to be singly, doubly, and triply metal-coordinated
–OH2 and –O surface groups. The protonation and deprotonation reactions of
surface –OH can be represented by
Mn –OHnv %
2 & Mn –OH
nv−1
+ H+ (Kn,1 ) (7.8a)
Mn –OHnv−1 %
& Mn –Onv−2 + H+ (Kn,2 ) (7.8b)
where n = 1, 2, 3 is the number of metal cations coordinated with surface –OH, and
v is the bond valence reaching the oxygen or hydroxyl. If one assumes the presence
of only one type of reactive surface group and if nv equals 1, the reactions (7.8)
simplify to the reactions (7.4). The intrinsic affinity constants Kn,i of the various
surface groups depend on the local configuration of the surface, i.e. the number of
coordinating cations, the valence of the cation, and the coordination number of
the central cations in the crystal structure. The difference between log Kn,1 and log
Kn,2 for proton binding on groups with the same surface configuration is so high
(about 14 log K units) that only one protonation reaction can take place for each
surface group within the normal pH range (3–11). Therefore, several Langmuir
isotherms can coexist, one for each surface site configuration, i.e. n = 1, 2, 3.
Schwarz and coworkers (Contescu et al., 1993a,b, 1995b) pictured the oxide
surface as consisting of oxo and hydroxo groups characterized by different proton
affinities. Using potentiometric titration data, they developed a method for calcu-
lating the proton affinity distribution of binding sites. In this way, the heterogeneity
of surface proton-binding sites was demonstrated. For the case of γ -Al2 O3 , it was
proposed that four categories of surface sites contribute to proton binding and sur-
face charge development between pH 3 and 11. They were identified with types I-a
(terminal, bound to tetrahedral Al), I-b (terminal, bound to octahedral Al), II-a
(bridging, bound to octahedral and tetrahedral Al), and III (triple-coordinated,
bound to octahedral Al) of surface hydroxyls. Type III are the most acidic sites,
while type I are the most basic ones. These surface groups react with solution
protons, depending on their log Ki and the solution pH, as follows:
[(Al Oh )3 –O]−0.5 + H+ %
& [(Al Oh )3 –OH]+0.5 , log KIII ∼ 2.5
[(Al Oh )–O–(AlTd )]−0.75 + H+ %
& [(Al Oh )–OH–(AlTd )]+0.25, log KII-a ∼ 4.1
−0.25 + %
[(AlTd )–OH] + H & [(AlTd )–OH2 ]+0.75 , log KI-a ∼ 6.8
[(Al Oh )–OH]−0.5 + H+ % & [(Al Oh )–OH2 ]+0.5 , log KI-b ∼ 9.8.
(7.9)
In Figure 7.2 the experimental proton adsorption isotherm of alumina is super-

imposed on the calculated proton affinity distribution, f(log K), and the contribu-
tions of the various binding sites are shown. Due to site heterogeneity, different
sites are present on the solid surface that are able to interact with either positive
or negative species from the impregnating solution, at any solution pH. Changing
Figure 7.2. Proton binding isotherm (proton consumption

vs pH) and the corresponding proton affinity distribution
[ f (log K) vs log K] for γ -Al2 O3 . Boxes assigned on basis
of local coordination of OH groups. (From Contescu et al.,
1995b.)
the pH results in variations in both number and type of surface sites with posi-
tive and negative charges. Depending on the solution pH, one or more types of
surface sites may contribute to adsorption of the catalyst precursor. Hence ad-
sorption equilibrium can be described by a combination of Langmuir isotherms,
each associated with one type of surface site contributing to adsorption (Contescu
et al., 1993a). This model predicts that even at the apparent PZC, some of the sites
are still charged. This explains adsorption data which contradict Brunelle’s model,
such as the adsorption of cations at pH < pHPZC , which in the case of γ -Al2 O3
is attributed to a negative charge of II-a sites at low pH (Contescu et al., 1995b).
In addition, geometrical constraints between the size and shape of the adsorbing
ions and the ensembles of charged surface sites required for adsorption may also
have to be allowed for, since these can be important factors that determine the
amounts of catalyst precursor adsorbed (Contescu et al., 1993a, 1995a).
7.1.2.c Ionic Strength

The ionic strength of the impregnating solution affects the adsorption capacity of
the support. It is defined by
1 2
I= Z i Ci (7.10)
2
where Zi is the valence and Ci the concentration of the individual ions. Ionic
strength affects the electric double-layer thickness and activity coefficient of pre-
cursor ions in the impregnating solution (Heise and Schwarz, 1986). It has been
demonstrated experimentally that the total uptake of metal precursor by the sup-
port decreases when the ionic strength of the impregnating solution increases
(Heise and Schwarz, 1986, 1987; Shah and Regalbuto, 1994). This can be explained
by considering that electrolyte ions modify the adsorption of metal precursor ions
by altering the charge distribution near the surface of support. According to the
Poisson–Boltzmann theory (Heise and Schwarz, 1987; Karpe and Ruckenstein,
1989), the surface charge creates an electrical potential whose extension into the
bulk solution is determined by the ionic strength and decreases with increas-
ing ionic strength. In other words, the electric field of the surface is increasingly
shielded with increasing ionic strength. Therefore the attractive force between
the oppositely charged surface and precursor ion diminishes, resulting in lower
precursor uptake. This is confirmed by the observation that univalent and diva-
lent compounds cause similar decreases in the amount of precursor adsorption
when the ionic strengths of the impregnating solutions are equivalent (Heise and
Schwarz, 1986).
7.1.2.d Precursor Speciation

The adsorption of metal complexes on the support can result from various types of
precursor–support interactions which are not just of electrostatic nature. In general
these include ion exchange or electrostatic adsorption (the immobilized species
is not altered), ligand exchange with surface hydroxyl groups (the coordination
Figure 7.3. Concentration of platinum complexes in

an aqueous solution of hexachloroplatinic acid as a
function of pH, regulated by addition of aqueous so-
lution of HCl or NaOH. (From Mang et al., 1993.)
sphere of the immobilized species is changed), and the formation of new chemical
compounds at the interface (the chemical identity of the immobilized species is
changed) (Westall, 1986; Schwarz et al., 1995). The interaction of metal precursor
with the support depends on its actual form in solution, i.e. its chemical specia-
tion. The latter is affected by the properties and concentration of the precursor,
the solution pH, the ionic strength, and the presence of added or extraneous
counterions (Schwarz, 1992). Metal ions in water are not bare, but hydrated. Thus
they can participate in exchange reactions where the coordinated water molecules
are exchanged for some preferred ligands. Hydrated multivalent metal ions can
in principle donate a larger number of protons than that corresponding to their
charge and can form anionic hydroxo–metal complexes.
For the case of aqueous solutions of chloroplatinic acid, Mang et al. (1993)
calculated the pH-dependent distribution of chloro–aqua and chloro–hydroxo
complexes of platinum(IV), when HCl and NaOH were used to adjust the pH. As
shown in Figure 7.3, when the pH increases the charge of the dominant species
changes from 0 to −2, while chlorine ligands are progressively replaced by H2 O
and OH− . It should be evident that the initial concentration of the precursor can
influence the equilibrium concentrations of the formed complexes.
Palladium ions in water containing chloride and/or ammonium ions form vari-
ous complex species with chloride and ammonium ligands, such as PdCl2− 4 , PdCl3
(H2 O)− , Pd(NH3 )2+4 , with electric charges ranging from −2 to +2 depending on
the solution pH (Contescu and Vass, 1987). Spielbauer et al. (1993) studied the
adsorption of palladium amino–aqua complexes on γ -Al2 O3 and SiO2 by UV spec-
troscopy, and showed that the speciation of palladium ions takes place according
to the following reactions, when pH adjustment was made by the use of ammonia:
[Pd(NH3 )n (H2 O)4−n ]2+ + H2 O %

& [Pd(NH3 )n−1 (H2 O)4−(n−1) ]2+ + NH3 .
(7.11)
The charge of all complexes was +2, while the number of NH3 ligands, n, increased
from 0 to 4 as the pH was increased. When adding an aqueous solution of NaOH,

amino–aqua and/or amino–hydroxo complexes are formed:
[Pd(NH3 )4−n (H2 O)n ]2+ + OH− %

& [Pd(NH3 )4−n (H2 O)n−1 (OH)]+ + H2 O.
(7.12)
7.1.2.e Coimpregnants
The presence of coimpregnants or extraneous ions in the impregnating solu-
tion can affect adsorption of the metal containing ions/complexes on the sup-
port through various mechanisms. The coimpregnants may be classified according
to the mechanism through which they affect interfacial phenomena (Heise and
Schwarz, 1987; Schwarz and Heise, 1990). The first class consists of inorganic
salts such as NaCl, NaNO3 , and CaCl2 , which affect ionic strength. The second
class includes inorganic acids and bases such as HCl, HNO3 , and NaOH, which
affect the pH of the system. These compounds can also partially dissolve the ox-
ide surface. The third class consists of compounds that compete with the metal
ion for adsorption sites. Although many chemical species can adsorb on the sur-
face, the strongest and most effective are those that contain hydroxyl, carboxyl,
and phosphoryl groups, such as acetates, citrates, and phosphates. Note that these
compounds also affect the pH and ionic strength of the solution.
Coimpregnants from the third class are commonly used to obtain nonuniform
catalyst distributions. Single component adsorption parameters of citric acid on
alumina are shown in Table 7.2. Large equilibrium adsorption constants have been
obtained (2070–19,530 l/mol), indicating strong adsorption. Surface saturation
coverage ranges from 1.43 to 3.35 µmol/m2 . Adsorption of other acids such as
acetic, lactic, oxalic and tartaric on alumina were also found to exhibit Langmuir
behavior (Engels et al., 1987; Jianguo et al., 1983).
During multicomponent adsorption, competition of the adsorbates for binding
sites takes place. Jianguo et al. (1983) showed that the amount of H2 PtCl6 adsorbed
on η-Al2 O3 was smaller in the presence of a coimpregnant. The decrease in amount
adsorbed as compared with the single-component H2 PtCl6 adsorption depended
not only on the concentration but also on the type of coimpregnant. Adsorbed
platinum in the presence of citric or tartaric acid was small, while in the presence
Table 7.2. Values of adsorption parameters of citric acid on γ -Al2 O3

in water solution.
Area K k+ ns te
(m2 /g) (l/mol) (l/mol · s) (µmol/g) (µmol/m2 ) (h) Reference
150 215 1.43 2 Shyr and Ernst, 1980
245 820 3.35 6 Jianguo et al., 1983a
220 2,070 444 2.02 2 Engels et al., 1987
78 19,530 7.97 209 2.68 1 Papageorgiou et al., 1996
a Solid was η-Al2 O3 .
of acetic, chloroacetic, or lactic acid it approached the values obtained by single-

component H2 PtCl6 adsorption. This is related to the fact that the first group
of acids exhibited strong single-component adsorption, while for the second the
adsorption was weak.
Competition for adsorption sites does not take place at low concentration of
coadsorbing ions. Xue et al. (1988) investigated the adsorption on alumina of
H2 PtCl6 and H2 IrCl6 , which show similar adsorption characteristics. The adsorbed
amount of a single metal from a bimetallic solution was the same as that from a
pure solution. This is because at low concentrations, the spacing between adsorbed
species on the surface is large and thus their interaction is negligible.
Papageorgiou et al. (1996), studied multicomponent adsorption of hexachloro-
platinic (1) and citric (2) acids on γ -Al2 O3 . It was found that a competitive Lang-
muir adsorption model (i.e. no interactions between the adsorbates, either in the
adsorbed or in the fluid phase)
dn1
= k1+ C1 (ns − n1 − n2 ) − k1− n1 (7.13a)
dt
dn2
= k2+ C2 (ns − n1 − n2 ) − k2− n2 (7.13b)
dt
was inconsistent with the measured data. Thus, an empirical model which took
account of solution effects for hexachloroplatinic acid and steric hinderance effects
for citric acid was formulated:
dn1
= k1+ C1 (ns1 − n1 ) − k1− n1 − k1sol C2 n1 (7.14a)
dt
dn2
= k2+ C2 ns2 − n2 − K2st n1 − k2− n2 . (7.14b)
dt
The introduction of the solution-effects term was based on the observation that
the amount of hexachloroplatinic acid desorbed when γ -Al2 O3 was subsequently
placed in a citric acid solution was in excess of that predicted by the single-
component model but less than that predicted by the competitive Langmuir model.
Also, the initial desorption rate of hexachloroplatinic acid was larger than that
computed from both models, suggesting that its desorption is enhanced by the
presence of citric acid in solution. The introduction of the term for steric hin-
derance provided by the adsorbed platinum was based on the observation that
the amount of citric acid adsorbed during multicomponent experiments was less
than that expected from the competitive Langmuir model. The expressions for
the multicomponent equilibrium isotherms including the new terms are
K1 C1e ns1
n1e = (7.15a)
1 + K1 C1e + K1sol C2e
K2 C2e
n2e = ns2 − K2st n1e (7.15b)
1 + K2 C2e
where K1 = k1+ /k1− , K2 = k2+ /k2− , and K1sol = k1sol /k1− . The two new parameters,
K1sol and K2st , were estimated by fitting the equilibrium data to be 735 l/mol and 3.18,
respectively. For the multicomponent adsorption rate constants, it was found that
the single-component value for hexachloroplatinic acid (2.46 l/mol · s) provided a
satisfactory representation of the transient multicomponent data. However, the
adsorption rate constant of citric acid had to be lowered to 0.2 l/mol · s in order to
achieve the best fit.
7.1.2.f Nature of the Solvent

So far our discussion has been limited to aqueous systems, since water is the most
common solvent used for catalyst impregnation. The solvent affects precursor
adsorption, because in the case of electrostatic interaction the stability of the
surface–ion pair depends on the dissociative power of the solvent. For solvents
with large dielectric constants, this power is large and results in a lower stability
of the surface–ion pair. Note that the dielectric constant of the solvent in the first
layer above the surface can be smaller than its bulk value, as in the case of water,
where the water dipoles in the first layer are aligned (Che and Bonneviot, 1988).
7.1.3 Surface Ionization Models

Surface ionization models characterize adsorption on oxide surfaces in terms of
chemical and electrostatic interactions. They are typically based on the assumption
that species bind to surface sites, and hence they are also called surface complex-
ation models. Depending on the depiction of the solution–support interface and
the species employed to describe the surface reactions, different models can be
obtained. These models and their applicability have been discussed in various re-
view papers (Westall and Hohl, 1980; Westall, 1986; Barrow and Bowden, 1987;
Haworth, 1990).
When contacting an oxide with an aqueous solution, reactions of the type
(7.4a,b) take place at the surface. They lead to the formation of protonated and
deprotonated surface sites (S–OH+ −
2 and S–O ), which are responsible for the sur-
face charge. In the absence of electrolyte or precursor ions, and neglecting the
contributions of H+ and OH− (Westall, 1986; Charmas et al., 1995), the surface
charge is given by
eNA
σ0 = ([S–OH+ −
2 ] − [S–O ]) (7.16)
NX
where σ0 has units of C/m2 , NA is Avogadro’s number, NX is the surface area of
the oxide per unit of solution volume (m2 of surface per m3 of solution), and e is
the elementary charge. In order to compute the concentration of surface sites, we
consider the mass action equations corresponding to reactions (7.4a,b),
[S–OH][H+s ]
Ka1 = + (7.17)
[S–OH2 ]
[S–O− ][Hs+ ]
Ka2 = . (7.18)
[S–OH]
The concentration of surface species, [H+

s ], is related to the bulk concentration
[H+ ] by the Boltzmann distribution
[H+ +
s ] = [H ] exp(−e'0 /kB T) (7.19)
where '0 is electrostatic potential at the surface. The exponential factor represents
the electrostatic energy required to bring a charged species from the solution bulk
to the adsorption surface. Using (7.19), equations (7.17) and (7.18) become
[S–OH][H+ ]
Ka1 = exp(−e'0 /kB T) (7.20)
[S–OH+
2]
[S–O− ][H+ ]
Ka2 = exp(−e'0 /kB T). (7.21)
[S–OH]
The material balance equation for surface sites is
NA
Ns = ([S–OH] + [S–O− ] + [S–OH+
2 ]) (7.22)
NX
where Ns is the density of surface sites (sites/m2 ). The equations above refer to
the water/oxide system. When considering the adsorption of catalytic precursors,
reactions of ions with surface sites must be included. This results in additional
equations, i.e. mass action laws related to these new surface reactions, and in
appropriate modifications of the equations above.
It is worth noting that the choice of the surface species is arbitrary and involves
the introduction of additional parameters which have to be determined by fitting
experimental data (Barrow and Bowden, 1987). A simple example is discussed
later in the context of the triple-layer model.
The surface complexation model described by the four equations (7.16), (7.20)–
(7.22) includes five unknowns ([S–OH], [S–O− ], [S–OH+ 2 ], σ0 , '0 ) and three pa-
rameters (Ka1 , Ka2 , Ns ), if the experimental pH is known. Thus, one more equation
is required. This is given by a relation between surface charge σ0 and surface po-
tential '0 , which is based on the electrostatic model employed to describe the
interface. Some models for this purpose are discussed next.
7.1.3.a Constant-Capacitance Model

With regard to adsorption of ions on a charged surface, we can distinguish bet-
ween specific and nonspecific adsorption. The first involves chemical bonds with
the surface active sites, which are specific to the ions involved. The second is
due to the electric potential of the surface and depends only on the ion charge.
The constant-capacitance, or single-layer, model (see Figure 7.4) assumes that all
specifically adsorbed ions are immobilized on the surface (plane 0) and contribute
to the charge σ0 together with the protonated and hydroxylated surface sites. All
nonspecific adsorbed ions are excluded from the 0 plane. The potential is related
linearly to surface charge by
σ0 = c'0 (7.23)
Figure 7.4. Electrostatic models for the surface–electrolyte-

solution interface. In each case, adsorbent bulk is to the left
of ordinate axis and solution to the right. The diagrams indi-
cate the mean planes to which individual classes of ions are
allocated along with the corresponding charges and show the
change in electrostatic potential with distance. C+ and A−
refer to specifically adsorbed cations and anions respectively.
where c is the capacitance. The latter is an unknown parameter which has to be

estimated by fitting experimental data. Its value is in principle valid only for the
species and concentrations for which the data were obtained.
7.1.3.b Diffuse-Layer Model

The electrostatic properties of a surface in contact with an aqueous phase can
be described by the diffuse-layer model. As shown in Figure 7.4, the interface
is regarded as consisting of two regions: an inner region and an outer diffuse

region (Gouy–Chapman layer). In the first, the ions are chemically bound to
the surface and are relatively immobile. In the second, they are bound only by
electrostatic interactions and undergo thermal motion. The distribution of ions in
the diffuse region is given by the Poisson–Boltzmann equation with appropriate
BCs. In particular, the following expression for the diffuse-layer charge is obtained
in the case of univalent electrolyte:
σd = −(8εr ε0 Rg TI)1/2 sinh(F'0 /2Rg T) (7.24)
where εr is the relative permittivity of the solvent, ε0 is the permittivity of vacuum,
I is the solution ionic strength, and F is the Faraday constant. Due to electroneu-
trality, the surface charge is given by σ0 = −σd . In the limit of low potential, the
hyperbolic sine function can be approximated by its argument, leading to
F
σ0 = (8εr ε0 Rg TI)1/2 '0 . (7.25)
2Rg T
The above expression is similar to equation (7.23) for the constant-capacitance
model. The difference is that the capacitance is now determined by theory.
7.1.3.c Basic Stern Model

This is a version of the basic Stern model (1924) as implemented by Bowden et al.
(1977). The protonated and deprotonated surface sites and the solvent ions H+
and OH− are assigned to the surface, where they contribute to the charge σ0 and
experience the potential '0 . Specifically adsorbed ions are assigned to a plane,
next to the surface, called the inner Helmholz plane (IHP), where they contribute
to the charge σIHP and experience the potential 'IHP . All nonspecifically adsorbed
ions are excluded from the 0 plane and the IHP, and are assigned to the diffuse
layer. As shown in Figure 7.4, the diffuse layer starts from a plane next to the IHP,
which is called outer Helmholz plane (OHP). This is the location closest to the
surface that the ions of the diffuse region can reach. The potentials of the IHP
and OHP are set equal, 'IHP = 'OHP , and hence the capacitance between them
is assumed to be infinite. The diffuse-layer charge σd can be calculated from (7.24)
where '0 is replaced by 'OHP . In the region between the 0 plane and the IHP,
constant capacitance c1 is assumed:
σ0 = c1 ('0 − 'IHP ). (7.26)
A charge balance equation for the IHP and an equation arising from electroneu-
trality are also required. These equations are discussed below in the context of
the triple-layer model.
7.1.3.d Triple-Layer Model

The triple-layer model can be regarded as an extended Stern model (Davis et al.,
1978; Davis and Leckie, 1978, 1980). It consists of two constant-capacitance layers
and a diffuse layer. The protonated and deprotonated surface sites and the H+ and
OH− ions are again located at the surface plane 0, and experience the potential
'0 . The IHP is separated from the surface by a region of constant capacitance c1
and contains specifically adsorbed ions at potential 'IHP . The IHP is separated
from the OHP by a region of capacitance c2 . The potential at the OHP is 'OHP and
the total diffuse layer charge is σd . The relationships between charge and potential
for the 0 plane, IHP, and OHP are
σ0 = c1 ('0 − 'IHP ) (7.27)

σIHP = c2 ('IHP − 'OHP ). (7.28)
The charge in the diffuse layer is calculated by equation (7.24), where '0 is re-
placed by 'OHP . The electroneutrality condition is given by
σ0 + σIHP + σOHP = 0 (7.29)
In a typical impregnating system for catalyst preparation, besides water and

the solid, several electrolytes are present. The adsorption of electrolyte ions is
considered to take place according to the site-binding model of Yates et al. (1974),
i.e., the adsorbed ions form “ion pairs” with charged surface sites. Therefore,
the electrolyte ions are considered as specifically adsorbed ions. The material
and charge balance equations now have to include these additional species. For
example, for a C+ A− electrolyte, the material balance equation for surface sites is
NA
Ns = ([S–OH] + [S–O− ] + [S–OH+ − + + −
2 ] + [S–O C ] + [S–OH2 A ]).
NX
(7.30)
The charges at the surface and at the IHP are given by

eNA
σ0 = ([S–OH+ + − − − +
2 ] + [S–OH2 A ] − [S–O ] − [S–O C ]) (7.31)
NX
eNA
σIHP = ([S–O− C+ ] − [S–OH+ −
2 A ]). (7.32)
NX
In addition to mass action equations for the surface sites (7.20)–(7.21), mass action
equations for the formation of surface complexes S–O− C+ and S–OH+ −
2 A have
to be included. If the surface reactions are described by
S–O− + C+ % − +
s & S–O C (KC ) (7.33a)
S–OH+ − % + −
2 + As & S–OH2 A (KA ) (7.33b)
(where subscript s denotes the surface), the mass action equations are
[S–O− C+ ]
KC = (7.34)
[S–O− ][C+ ] exp(−e' IHP /kB T)
[S–OH+ −
2A ]
KA = . (7.35)
[S–OH+ −
2 ][A ] exp(+e'IHP /kB T)
The entire set of equations (7.20), (7.21), (7.24), (7.27)–(7.32), (7.34), (7.35) can be
solved numerically at any given pH and electrolyte concentration. The parameters
are Ka1 , Ka2 , KA , KC , Ns , NX , c1 , and c2 , while the unknowns are [S–O− C+ ],

[S–OH+ − − +
2 A ], [S–OH], [S–O ], [S–OH2 ], σ0 , σIHP , σd , '0 , 'IHP , 'OHP . Some of
these parameters (Ns , NX ) can be determined by independent means: the den-
sity of reactive sites through surface titration (such as tritium exchange), and the
surface area of the oxide by gas adsorption (Westall and Hohl, 1980). The other
parameters are adjustable and can be estimated only through surface hydrolysis
experiments. The model can incorporate more than one reaction to describe the
adsorption of an ion on the surface, and a solute can give a large number of anions
and cations according to its chemical speciation. Both lead to a larger number of
adjustable parameters. Using the triple-layer model in its original form or with
modifications, adsorption of various metal species has been described (Hachiya
et al., 1984; Hayes and Leckie, 1986; Mang et al., 1993; Spielbauer et al., 1993). A
revised version of this model has also been presented by Righetto et al. (1995).
7.1.3.e Four-Layer model

The four-layer model was introduced by Bowden and Barrow (Bowden et al., 1980;
Bolan and Barrow, 1984; Barrow and Bowden, 1987) and can be regarded as an
extension of the triple-layer model. A fourth plane, a, is introduced between the 0
plane and the IHP. The specifically adsorbed ions are located at the IHP or at the a
plane, depending upon the strength of adsorption. The ions with stronger affinity
to the surface are closer to it, i.e. at the a plane, while those with lower affinity
are further away at the IHP. Charmas et al. (1995), in their implementation of the
four-layer model, assigned cations to the a plane and anions to the IHP, because of
their different size. The final equations are similar to those used for the triple-layer
model, and are not discussed here in detail.
7.2 Simultaneous Diffusion and Adsorption in Pellets

Nonuniform catalyst profiles usually arise from the interaction of intrapellet flow,
diffusion, and interfacial phenomena, depending upon the particular system and
the impregnation process employed. In the latter, the support contacted with the
impregnating solution is either dry (dry impregnation) or filled with the solvent
(wet impregnation). In the first period of dry impregnation, i.e. imbibition, due to
capillary forces, the solution fills the pores of the support and hence the solute is
transported primarily by convective flow. After imbibition in dry impregnation,
or during the entire course of wet impregnation, transport of solute is by diffu-
sion. Adsorption of the solute by the pore walls occurs simultaneously with solute
transport. The drying process following impregnation can also affect the final cat-
alyst distribution. The total catalyst deposited on the internal surface of the pellet
comprises the precursor adsorbed during impregnation and that precipitated from
the unadsorbed solute during drying. The relative contributions of the two sources
to local catalyst loading can be widely different, depending on the support and
precursor properties, as well as on the catalyst preparation conditions.
The most common strategy for preparing nonuniformly distributed catalyst
pellets is to first realize the desired distribution of the adsorbed precursor, and
then dry the pellet, trying not to modify it. Therefore, in this section we focus on
the impregnation process, both theoretically and experimentally, with the aim of
determining the distribution of adsorbed precursor within the pellet. We conclude
by discussing several studies which address nonuniform catalyst distributions from
both experimental and modeling viewpoints.
7.2.1 Theoretical Studies

Various models of wet and dry impregnation have been developed (cf. Melo et al.,
1980a; Lee and Aris, 1985). Intraparticle convective flow is due to capillary action,
and its description is based on the Washburn equation or Darcy’s law. Diffusion
is typically described by Fick’s law. Different variations of Langmuir isotherm
models have been considered for the uptake of solute on the support, and a few
investigators have also used surface ionization models.
7.2.1.a Dry Impregnation

For a porous spherical pellet the mass balances of impregnants during dry impreg-
nation are

∂Cp,i ∂ni υ ∂Cp,i De,i ∂ 2 ∂Cp,i
ε + ρs = + x (7.36)
∂t ∂t 4π x 2 ∂ x x2 ∂ x ∂x
with BCs
∂Cp,i
x = R: υ(Cb,i − Cp,i ) = 4π R2 De,i (7.37a)
∂x
∂Cp,i
x = rf : ρs υni = 4πrf2 ε De,i (7.37b)
∂x
and initial conditions
t = 0: Cp,i (x, 0) = ni (x, 0) = 0 (7.37c)
where Cp,i is the concentration of the ith solute in the pores, ρs is the pellet density,
and υ is the volumetric flow rate of imbibition, related to the position rf of the
advancing front by
drf
υ = −4πrf2 ε . (7.38)
dt
The derivative ∂ni /∂t represents solute uptake by the support and may be obtained
from the Langmuir adsorption equation (7.1) (for single component adsorption)
or (7.13) (for two-component competitive adsorption). In the case where adsorp-
tion is fast compared to the transport processes, the assumption of local adsorption
equilibrium can be used. Accordingly, the concentrations in the adsorbed phase
can be obtained directly from the concentrations in the pores through an adsorp-
tion model, such as for example the Langmuir isotherm equation (7.2). Therefore,
by taking the adsorption isotherm in the general form
ni = fi (Cp,1 , Cp,2 , . . . , Cp,N ) (7.39)

the expression for ∂ni /∂t in equation (7.36) becomes
∂ni N
∂ fi ∂Cp,k
= (7.40)
∂t k= 1
∂Cp,k ∂t
where N is the number of adsorbates.

In the imbibition stage of dry impregnation, the impregnating solution fills the
pores of the support. Convective transport dominates over diffusion. The vol-
umetric flow rate of imbibition, υ, and the position rf of the imbibed liquid at
time t are computed through an appropriate hydrodynamic model. Two com-
mon approaches for this are based on Darcy’s law (Lee and Aris, 1985) and on
the Washburn equation (Vincent and Merrill, 1974). In particular, Lee and Aris
(1985) derived equations for the front position and velocity under two different
conditions: where the air occupying the pore space before penetration was ei-
ther trapped in the pellet and compressed by the liquid front, or where it was
released from the pellet. After the liquid front reaches the center of the pellet, i.e.
rf = 0, imbibition ceases and υ = 0. From now on, diffusion is the only transport
mechanism. In addition, depletion of solute in the external solution must be taken
into account. The evolution of the system after imbibition is discussed later in the
context of wet impregnation.
We now consider catalyst deposition during imbibition. In this period, dif-
fusive transport may be neglected because convective transport dominates. For
a one-component system, equations (7.36)–(7.37) can be simplified for the case
of plug flow into a single pore (valid for slab shape pellets) to analyze the time-
dependent flow of impregnating solution and dispersal of the impregnant (Vincent
and Merrill, 1974):

∂Cp,i ∂Cp,i
ε +v = −ap Wi (7.41)
∂t ∂z
∂ni (7.42)
ρs = ap Wi
∂t
z = 0: Cp,i (0, t) = Cb,i (7.43a)
t = 0: Cp,i (z, 0) = ni (z, 0) = 0 (7.43b)
where ap is the pore surface area per unit volume, v is the velocity of the advancing
front computed using the Washburn equation, and Wi is the mass flux of impreg-
nants from liquid to the pore wall. Two situations were investigated for the solute
mass flux Wi . In the first, the rate was controlled by mass transfer resistance across
the liquid–solid interface, and in the second, by adsorption kinetics. Adsorption
was described by the Langmuir model. Both cases were shown to yield similar
eggshell impregnation profiles in systems where the uptake process was suffi-
ciently fast. Moreover, in these cases it was possible to control the impregnation
depth by adjusting the adsorption capacity.
For the same limiting cases, Lee (1984) developed analytical expressions for
the catalyst distribution at the end of the imbibition period, considering both the
precursor occluded in the pore and that adsorbed on the pore walls. This is the
impregnation profile that would be obtained following imbibition with fast drying.
An important extension included consideration of different one-dimensional pel-
let shapes (slab, cylinder, sphere) by allowing for the effect of geometry on the
imbibition front velocity v in equation (7.41). For this purpose, a modification of
the Washburn equation, developed by Komiyama et al. (1980), was used. It was
found that the obtained nonuniform catalyst distributions degenerate to uniform
ones only when sufficient time (of the order of days) is allowed for the equilibra-
tion between the occluded liquid and the adsorbed phase. Catalyst distribution in
a pellet can also be influenced by adjusting imbibition conditions so that only par-
tial penetration of the pores is achieved. Since convective transport of solution is
due to capillary forces, critical parameters that affect the fraction of pore volume
filled with liquid are solution viscosity, surface tension, support geometry, average
pore size, and immersion time (Zhang and Schwarz, 1992).
Coimpregnation of pellets with catalyst precursor and coingredients can be
used to manipulate intrapellet distribution of the active material. The model of
Vincent and Merrill for imbibition was extended to a multicomponent system by
Kulkarni et al. (1981). A competitive Langmuir adsorption model was considered.
It was shown that when the equilibrium fractional coverage for a given concen-
tration of precursor in the liquid phase is substantially smaller in the presence of a
coimpregnant, it is possible for the precursor profile to exhibit a maximum within
the pore. This condition is equivalent to stronger adsorption of the coimpregnant
than of the catalyst precursor.
Analytical solutions for catalyst distribution at the end of the imbibition period
were obtained by Lee (1985). The model was based on competitive Langmuir
adsorption and the modified Washburn equation mentioned above for the front
velocity. Since the advantage of utilizing a coimpregnant is best realized when
adsorption is the controlling step, mass transfer resistance at the pore wall was
neglected. Egg-white distributions were obtained if both impregnants adsorbed
strongly, but the coimpregnant (species 2) adsorptivity was larger (i.e. K1 was large
and K2 > K1 ) and the rate of its adsorption was higher (k+ +
2 > k1 ) than that for the
catalyst precursor. The catalyst layer’s location and thickness can be manipulated
by adjusting the bulk concentrations of the two impregnants. The location of the
catalyst layer depends also on the pellet shape, because the latter affects the ve-
locity of the penetrating liquid front. This occurs because the liquid front moves
faster, as the pellet center is approached, in a spherical than in a cylindrical pellet;
the velocity is independent of position in a slab pellet. Under identical preparation
conditions, a slablike pellet has its shell closer to the center than a cylindrical or
spherical pellet.
In an extensive analysis of catalyst impregnation, Lee and Aris (1985) consid-
ered various one- and two-component systems with both finite and infinite adsorp-
tion rates. The latter implies instantaneous adsorption, i.e., equilibrium prevails at
any time and location throughout the wetted portion of the pellet. Under this
condition, the concentration profiles take the form of shocks moving in time. For
two-component impregnation, three regions, each with a constant composition,
are found. The more strongly adsorbed coimpregnant (species 2) is present only in
Figure 7.5. Profiles of pore fluid (Cp ) and adsorbed

(n) concentrations from two-component mixture im-
bibition with instantaneous adsorption. Component 2
is more strongly adsorbed. (From Lee and Aris, 1985.)
the outer region, as illustrated in Figure 7.5. A portion of the less strongly adsorbed
precursor (species 1) passes through this region and penetrates deeper into the
pellet, where it faces no competition for the adsorption sites. Thus, the surface
concentration of adsorbed component 1 in this egg-white region is greater than
that in the eggshell region. The highly concentrated component 1 adsorbed in the
egg-white region is later washed off by the oncoming component 2, which causes
the overshoot in the concentration profile of component 1. As time progresses, the
band containing component 1 moves towards the interior of the pellet. When the
center is reached, then the egg-yolk distribution is obtained.
In the case of finite adsorption rate, the sharp concentration profiles become
blurred. Both the height and location of the catalyst layer are affected. As the rel-
ative rate of adsorption of the coimpregnant decreases, it becomes less effective in
pushing the precursor into a narrow subsurface layer. However, if the rheological
or interfacial properties of the impregnating solution are manipulated so as to
slow down the liquid imbibition rate, nonuniformities in the catalyst distribution
can be preserved.
Not only the adsorption rate but also the desorption rate affects the peak. If
the desorption rate of the precursor is decreased, its adsorbed part is washed off
slowly, smoothing the peak. Furthermore, the diffusion transport of solutes during
imbibition, superimposed on finite rates of adsorption, intensifies the blurring of
the otherwise sharp concentration profiles.
7.2.1.b Wet Impregnation

In wet impregnation the porous pellets, filled with pure solvent, are immersed in
the impregnating solution. Therefore no convective solute transport takes place
in the pores. Hence equation (7.36), with the convective flow term neglected, can
be used to describe the mass balance of the solutes:

∂Cp,i ∂ni De,i ∂ ∂Cp,i
ε + ρs = 2 x2 . (7.44)
∂t ∂t x ∂x ∂x
In addition, depletion of solutes in the external solution must be taken into
account:

dCb,i 2 ∂Cp,i
Vb = −Np De,i 4π R (7.45)
dt ∂ x x=R
where Vb is the volume of the external bulk fluid, Cb,i is the concentration of the
ith solute in the external bulk fluid, and Np is the number of spherical pellets. If
there are no external transport limitations, the BCs are
∂Cp,i
x = 0: =0 (7.46a)
∂x
x = R: Cp,i = Cb,i . (7.46b)
The initial conditions
t = 0: Cb,i = Cb 0,i , Cp,i (x, 0) = ni (x, 0) = 0 (7.47)
refer to coimpregnation, where precursor and coimpregnants are introduced at

the beginning of the process. The same equations (7.44)–(7.46) can be used to
describe the development of the catalyst distribution for dry impregnation after
the imbibition process is complete. Different initial conditions for Cp,i and ni ,
however, have to be considered. These are given by the solution of the imbibition
model discussed in the previous section.
Do (1985) studied one-component impregnation, where adsorption was de-
scribed by the Langmuir isotherm. It was shown that when the impregnation is
diffusion-controlled, a sharp eggshell solute distribution is obtained, particularly
when there is no desorption of the active species, so that there is no chance for
redistribution. The sharp front of the eggshell profile where the adsorption process
takes place moves towards the pellet center as time increases. A higher adsorption
capacity of the support results in a decrease of the front velocity and in an increase
of the overall loading, but it does not have a significant effect on the form of the
solute profile. When the solute’s adsorption is slower, or its diffusivity is larger,
it penetrates deeper into the support, yielding a smoother distribution front (Do,
1985; Vazquez et al., 1993). Similarly, in dry impregnation, the front of the eggshell
profile obtained after imbibition is complete, moves towards the center of the pel-
let with increasing time (Lee and Aris, 1985). For slow adsorption, imbibition can
yield a degenerate eggshell distribution with smoothly decreasing profile towards
the pellet center. By either prolonging the impregnation time after imbibition or
using wet impregnation with careful control of the impregnation time, a sharper
eggshell distribution can be produced. This occurs because diffusive transport is
much slower than convective transport. For the same reason, dry impregnation
yields a thicker catalyst shell than wet impregnation, but this difference becomes
less noticeable with longer impregnation time.
Figure 7.6. Comparison of adsorbed so-

lute distribution between wet impregna-
tion (solid lines) and dry impregnation
(dashed lines): profile 1, 1 min; profile
2, 5 min; and profile 3, 10 min. Impreg-
nation conditions Cb0,1 = 0.91 × 10−2 M,
Cb0,2 = 10Cb0,1 , ns = 1.88 × 10−6 mol/m2 ,
K1 = 1.83 × 103 mol/l, K2 = 1.5K1 , k1+ =
21.11 l/mol · s, k2+ = 1.5k1+ , De,1 = 10−5
cm2 /s, De,2 = 0.1De,1 . (From Lee and
Aris, 1985.)
In the earlier part of this book, we have seen many instances where egg-white
catalysts provide optimal performance. It is useful to describe here the general con-
ditions which favor the formation of such distributions. The egg-white distribution
obtained by two-component imbibition, as shown in Figure 7.5, is the result of com-
petition between adsorption and convective transport. Once imbibition is over, the
catalyst deposition becomes controlled by the interaction between adsorption and
diffusion. This may either destroy or enhance the egg-whiteness of the distribution.
Lee and Aris (1985) investigated different scenarios. For example, if the two com-
ponents have same diffusivity and they are both weakly adsorbable, then the egg-
white characteristic is not pronounced, and is quickly wiped out. However, when
the precursor has larger diffusivity, egg-whiteness is modified but not destroyed. If
in addition, the adsorption of both components is sufficiently strong, egg-whiteness
is enhanced as the impregnation time is extended. As in one-component impreg-
nation, the difference between dry and wet impregnation diminishes as the im-
pregnation time is prolonged, as illustrated in Figure 7.6. Stronger adsorption of
the coimpregnant is critical in obtaining egg-white catalysts. It can effectively push
the precursor into a narrow, sharp, and concentrated band even when both com-
ponents have the same diffusivity. Larger diffusivity of the precursor makes the
overshoot of the egg-white distribution broader and higher. The adsorption ca-
pacities of the impregnants can also affect the catalyst profile. If they are small, the
catalyst distribution evolves quickly from egg-white to egg-yolk and then becomes
uniform. During this evolution, the peak is never very pronounced.
Lee and Aris (1985) have summarized the conclusions of their extensive sim-
ulations in the form of quantitative criteria for obtaining distinct egg-white or
egg-yolk catalysts, in the case of purely competitive Langmuir adsorption:
1. > 103 l/mol, K2 > K1 .

K1 ∼
2. > 10 l/mol · s, k2+ ≥ k+ .
k+ ∼
1 1
3. k1− > 10−2 s−1 .
∼
4. De,1 ≥ De,2 if K2 K1 ; De,1 > De,2 if K2 = K1 .

5. ρs ns,2 /εCb0,2 > 1.
6. Careful control of impregnation time is required.
7.2.1.c Effects of Electrokinetic and Ionic Dissociation Phenomena

The models discussed in previous subsections do not explicitly allow for the in-
fluence of electrokinetic transport and ionic dissociation on catalyst distribution.
Ruckenstein and Karpe (1989) (see also Karpe and Ruckenstein, 1989) conducted
a detailed theoretical study of wet impregnation taking these phenomena into ac-
count. A surface ionization model (see section 7.1.3) was used for ion adsorption.
Chloroplatinic acid adsorption on γ -alumina was considered, in the presence of
NaI to adjust the ionic strength, HI and NaOH to vary the pH, and NaNO3 , NaCl,
and sodium citrate as competing coimpregnants. The axial flux inside the pores was
given by the Nernst–Planck equation, which combines Fickian diffusion with ionic
migration. The radial potential distribution within the pores was approximated by
the Poisson–Boltzmann equation. Ionic fluxes are influenced by electrical poten-
tial due to surface charge, which in turn is affected by ionic strength, pH and
adsorbed ions of the impregnating solution.
The effects of pH and ionic strength of the impregnating solution, nature of the
support, and coimpregnating species on the Pt distribution profile were examined
within this framework. It was found that for one-component impregnation (i.e. no
competing coimpregnant), the axial distribution decreases from the surface to the
interior of the pellet. When ionic strength increases, this distribution decreases
in both height and depth. This is a consequence of the decreased adsorptivity of
Pt due to increased ionic strength of the solution. Increase of external pH first
increases the extent of adsorption (and therefore the height and depth of the
adsorbed platinum distribution), which then passes through a maximum at pH
2.7, and eventually decreases. A comparison of this model with the more common
model based on Fickian diffusion and Langmuir adsorption (where both electroki-
netic transport and ionic dissociation are neglected; see section 7.2.1.b) and with
the case where only electrokinetic transport is neglected (i.e., Fickian diffusion is
the sole transport mechanism) is shown in Figure 7.7. These distributions are ob-
tained using sodium citrate as the competing agent. It is clear that predictions of the
three models are quite different. Electrokinetic phenomena affect the transport
processes, because electrical potential in the axial direction increases the flux of
coions and decreases the flux of counterions. The diffusion of protons and hydrox-
yls from the impregnating solution, coupled with the protonation–deprotonation
equilibria of hydroxyl groups of the surface, can give rise to a nonuniform pH pro-
file inside the pore, which affects not only ionic migration but also ion adsorption.
7.2.1.d Effect of Drying Conditions

The drying process following impregnation can alter catalyst distribution obtained
at the end of the impregnation step. Lee and Aris (1985) investigated catalyst
Figure 7.7. Model prediction of the effect of coimpreg-

nation on the platinum distribution profile with cit-
rate ion as coimpregnant. Solid curves: electrokinetic
and ionic effects considered; dotted curves: electroki-
netic effects neglected; dashed curves: electrokinetic
and ionic dissociation effects neglected. Curve a: ad-
sorbed impregnant in the absence of coimpregnation;
curves b: adsorbed impregnant during coimpregnation;
curves c: adsorbed coimpregnant. Chloroplatinic acid
0.5 × 10−3 M, sodium citrate 0.5 × 10−3 M, pH0 = 3.3,
time = 10 h, I0 = 10−3 M. (From Ruckenstein and
Karpe, 1989.)
redistribution during drying, using a simplified model of a capillary tube where

evaporation takes place at the open end of the tube. The liquid front moves from
the pore interior to the external surface, as expected in the case of slow drying
(see section 7.2.2.c). The model adopted is similar to that used for dry impregna-
tion, i.e. equations (7.36)–(7.38), where now the front velocity has the opposite
direction and is controlled by the drying rate. Solute precipitation and film flow
were neglected, while several initial solute distributions were considered. They
were altered during drying as a result of the interaction between adsorption and
convective and diffusive transport. It was shown that drying has little effect on
catalyst distribution in the case of strong adsorption. On the other hand, if ad-
sorption is weak, drying has a smearing effect on eggshell and egg-white catalyst
profiles developed at the end of the impregnation step. In this case, diffusion and
Figure 7.8. Model prediction of the effects of drying on

the Rh impregnation profiles across the wall of γ -Al2 O3
monoliths prepared by coimpregnation with RhCl3 and
HF. The curves are for adsorbed rhodium. Preparation
conditions: 3.8 × 10−5 mol/cm3 RhCl3 ; Cat. B: 3.3 M
HF; Cat. C: 6.6 M HF; Cat. D: 9.9 M HF. (From Hepburn
et al., 1991a.)
flow bring the solutes back into a region of substantial adsorption near the pore
mouth. However, the shape of the distribution can be preserved if the diffusivity
of the precursor is larger than that of the coimpregnant, because the former moves
inward faster and reaches available adsorption sites before the latter.
Hepburn et al. (1991a) studied catalyst redistribution during drying for the case
where evaporation takes place at an interface receding towards the pellet interior,
which is expected to occur in the case of fast drying (see section 7.2.2.c). The
physicochemical parameters used were representative of γ -Al2 O3 coimpregnated
with RhCl3 and HF, so as to realize an egg-white Rh distribution. The results
indicated that drying causes a buildup of RhCl3 and HF at the shrinking liquid front
where evaporation occurs. As this front moves towards the center of the support,
the increase in RhCl3 and HF concentrations in the liquid causes a competition
for unoccupied adsorption sites. RhCl3 is pushed ahead of HF because the latter
adsorbs more strongly than the former. As a result, after drying, the Rh band is
displaced towards the interior and also becomes less sharp than the Rh profile
before drying, as shown in Figure 7.8.
7.2.2 Experimental Studies

The catalyst precursor deposited on the surfaces of the pores of a pellet after
impregnation is composed of two parts: the first is solute adsorbed on the sup-
port during the impregnation process, and the second is from precipitation of
unadsorbed solute from the occluded pore liquid during the drying process. In
this subsection, we discuss the role of both these processes in determining the
final catalyst distribution through various experimental studies reported in the

literature. These include single- and multicomponent impregnation as well as the
effects of drying conditions.
7.2.2.a Single-Component Impregnation

When only the catalyst precursor is present during impregnation, and in the case
where drying does not affect distribution of the catalytic material, the relative
rates of transport and adsorption determine the distribution obtained. It has been
observed experimentally for various systems that solute transport faster than
adsorption yields relatively uniform distributions, while adsorption faster than
transport yields eggshell distributions (Chen and Anderson, 1976; Summers and
Ausen, 1978; Nicolescu et al., 1990). For intermediate situations, various types
of distributions are obtained. If the adsorbed species can desorb at a reasonable
rate, redistribution can occur by desorption and migration due to diffusion in
the pores. This can ultimately give a uniform catalyst distribution (Maatman and
Prater, 1957). Since the strength of adsorption is directly related to the adsorption
rate (K = k+ /k− ), large desorption rates are typically characteristic of weak and
not of strong adsorption. We now analyze the parameters that affect adsorption
and transport rates, and hence also influence the catalyst distribution.
The nature of the precursor strongly affects the active-element adsorption
characteristics. Summers and Ausen (1978) studied adsorption of various metal-
ammonia–chloride complexes and demonstrated that they had different adsorp-
tion characteristics depending on the number of chlorine and ammonium ions. In
the case of ammonium paratungstate, (NH4 )10 W12 O41 , a variety of species con-
taining one, six, and twelve tungsten atoms can be present in aqueous solution.
High tungsten salt concentration and low pH favor the existence of twelve-atom
species, which are large and interact strongly with alumina. As a result, they tend to
block the pores at the exterior of the pellet, resulting in eggshell profiles which per-
sist even after long impregnation times (Maitra et al., 1986). Moreover, depending
on whether the precursor is an acidic, basic, or neutral compound, its presence
affects the solution pH and through this also the adsorption characteristics (Roth
and Reichard, 1972; Chen and Anderson, 1973).
The type of solvent used in impregnation also plays an important role. For a
given precursor–support system, adsorption isotherms, saturation coverage, and
adsorption kinetics are different in aqueous and in organic impregnating solu-
tions (Hanika et al., 1982; Machek et al., 1983a,b). The nature of the solvent can
also affect the catalyst distribution through its volatility characteristics (Burch and
Hayes, 1997). A volatile solvent can result in faster drying of the pellet after im-
pregnation, which, as discussed below, can greatly affect the catalyst distribution.
Impregnation without solvent has also been attempted. In particular, Soled et al.
(1995) impregnated silica pellets with molten cobalt nitrate. The melt penetrated
the pellets slowly due to its high viscosity, finally yielding eggshell catalysts.
Using dry or wet impregnation leads to different distributions. For long im-
pregnation times this difference decreases, since the contribution of imbibition
to the final distribution decreases. For weak adsorption, Cervello et al. (1976)
and Melo et al. (1980b) found that dry impregnation yields a catalyst with higher
loading and more uniformity than wet impregnation. For CuCl2 adsorption on γ -
Al2 O3 , Ott and Baiker (1983) showed that the penetration depth obtained from
dry impregnation was larger than from wet impregnation. Summers and Ausen
(1978) observed similar trends, but the difference was less significant because the
impregnation time was long.
The concentration of impregnating solution and the impregnation time deter-
mine the thickness of the shell in the case of strong adsorption. It has been observed
for various systems that extended contact with the bulk impregnating solution al-
lows the solute to diffuse into the support and enlarge the eggshell (Maatman and
Prater, 1957; Harriott, 1969; Santacesaria et al., 1977c; Baiker and Holstein, 1983;
Lee and Aris, 1985; Nicolescu et al., 1990). For Ni(NO3 )2 and Ba(NO3 )2 impreg-
nation of alumina, which is characterized by weak adsorption, wet impregnation
with short contact time, or low concentration of the impregnant, produces catalysts
with decreasing concentration profile towards the pellet center. Increasing any of
the above operating parameters smooths catalyst nonuniformities (Melo et al.,
1980b). Goula et al. (1992b) studied the impregnation of ammonium heptamolyb-
date on γ -Al2 O3 . During dry impregnation they observed that with increasing
precursor concentration, less uniform distributions were obtained. Since the op-
posite trend was expected for weak adsorption, this behavior was attributed to
a pH effect. It was confirmed that by independently increasing the pH, keeping
constant precursor concentration, more uniform distributions were obtained. In
addition, the effect of impregnation time was investigated when imbibition was
incomplete. It was found that longer impregnation times lead to smoother eggshell
profiles when imbibition was incomplete, while further increase after imbibition
was complete resulted in sharpening the eggshell distribution. This is because ad-
sorption is faster than diffusion, thus favoring precursor adsorption at the pore
mouth.
Goula et al. (1992b) did not observe any effect of the impregnating solution vol-
ume on the catalyst distribution profiles, which is reasonable for weakly adsorbing
species, since no significant precursor depletion occurs in the bulk. For the case of
strong adsorption, though, the volume of the solution would be expected to have
a similar effect as its concentration, since it would affect the amount of precursor
available for adsorption. Even for long impregnation times, since desorption is
slow, the precursor is distributed nonuniformly if the amount in solution is less
than that required to saturate, whereas if there is an adequate amount of solute,
a uniform distribution can be obtained (Maatman, 1959).
The impregnating solution’s pH and ionic strength have an indirect but signifi-
cant effect on adsorption characteristics of the precursor. By altering the pH in the
direction which favors adsorption of the impregnating species, sharper eggshell
distributions can be obtained (Maitra et al., 1986; Heise and Schwarz, 1985; Goula
et al., 1992b). For the system H2 PtCl6 /γ -Al2 O3 , when the ionic strength of the im-
pregnating solution was increased by addition of simple inorganic salts, Heise
and Schwarz (1986) found that the penetration depth of the eggshell distribution
was not affected, but the adsorbed precursor concentration decreased and the
Figure 7.9. Experimental platinum adsorp-

tion profiles produced by impregnating
γ -alumina pellets with 0.0025 M hexachloro-
platinic acid and various sodium nitrate con-
centrations: (a) 0.0 M, (b) 0.005 M, and
(c) 0.025 M. (From Heise and Schwarz,
1986.)
distribution front smoothed, as shown in Figure 7.9 for γ -Al2 O3 impregnation

with hexachloroplatinic acid in the presence of sodium citrate.
For ammonium heptamolybdate [(NH4 )6 Mo7 O24 ] impregnation of γ -Al2 O3
extrudates, Goula et al. (1992b) smoothed eggshell catalyst profiles by decreasing
the impregnation temperature and doping the support with fluoride ions. Using
these methods, the pHPZC could be decreased and moved closer to the impregna-
tion pH, thus reducing the concentration of adsorption sites.
7.2.2.b Multicomponent Impregnation

Most multicomponent impregnation studies concern platinum supported on γ -
Al2 O3 . Shyr and Ernst (1980) studied the influence of a large number of acids
and salts, used as coimpregnants with H2 PtCl6 , on the distribution of Pt within
γ -Al2 O3 pellets, and obtained the profiles shown in Figure 7.10. Among others,
these include eggshell, egg-white, and egg-yolk distributions. The ability to pre-
pare these distributions for Pt provides the motivation for investigating other
Figure 7.10. Types of Pt profiles obtained in coimpreg-

nation experiments. (From Shyr and Ernst, 1980.)
multicomponent impregnation systems, in order to identify conditions for obtain-

ing a desired catalyst distribution. The underlying principles have been discussed
in section 7.2.1.
If HCl is used as coimpregnant along with chloroplatinic acid, a uniform plat-
inum distribution is obtained (Maatman, 1959; Van den Berg and Rijnten, 1979;
De Miguel et al., 1984; Scelza et al., 1986). When dibasic organic acids (for exam-
ple citric or oxalic acid) are used as competitors, platinum is driven into the pellet
interior, forming a subsurface layer with controlled location and width (Michalko,
1966b; Hoekstra, 1968; Summers and Hegedus, 1978; Becker and Nuttall, 1979;
Li et al., 1985; Papageorgiou et al., 1996). HF has also been used as coimpregnant
to prepare catalysts with subsurface layers of Pt (Hepburn et al., 1989a; Shyr and
Ernst, 1980), Pd (Summers and Hegedus, 1978; Hepburn et al., 1989a), and Rh
(Hegedus et al., 1979b; Hepburn et al., 1991a,c).
Hepburn et al. (1989c) prepared eggshell and egg-white Rh/γ -Al2 O3 catalysts
using hydrofluoric acid, hydrochloric acid, citric acid, and their sodium salts as
coimpregnants. It was shown that rhodium can be driven beneath the external
surface of the support more effectively by using coimpregnants with a low degree
of dissociation in aqueous solution. As such, HF and citric acid result in egg-
white distributions, while the other coimpregnants do not significantly influence
the distribution type.
When adjusting the impregnating conditions to control the catalyst-layer
location, the catalyst loading also usually changes. This occurs through a change
of the width or local loading in the catalyst layer and/or the region between the
external pellet surface and the catalyst layer (Hegedus et al., 1979b; Lee and Aris,
1985; Hepburn et al., 1991c; Papageorgiou et al., 1996). In catalyst preparation
technology, it is desirable to control catalyst loading and location separately. This
can be accomplished by using sequential impregnation as discussed below. For

coimpregnation, a method that has been proposed to keep total loading constant
is to form macromolecules which contain the metal and adsorb at the pellet exter-
nal surface. This is done by introducing ethyl silicate in the impregnating solution
so that (–M–O–Si–)n structures are formed. Once a certain amount of macro-
molecules adheres over the pellet surface, no additional molecules adsorb. This
results in a constant loading of metal ions in the catalyst pellet. Subsequently, the
adsorbed macromolecules decompose on the pellet surface, and the metal ions
diffuse into the pellet. In this case, the catalyst location is controlled by the im-
pregnation time, and the catalyst loading by the concentration of precursor in the
impregnating solution (Fujiyama et al., 1987; Otsuka et al., 1987).
In coimpregnation, two key parameters for controlling the catalyst distribution
are impregnation time and competitor concentration. In general, it is observed that
by increasing impregnation time the catalyst layer of an egg-white pellet is pushed
deeper inward, yielding egg-yolk distributions. At long impregnation times, these
subsurface cores can be washed out by back diffusion, producing uniform distri-
butions with lower local loading (Becker and Nuttall, 1979; Hegedus et al., 1979b;
Shyr and Ernst, 1980; Hepburn et al., 1991c; Papageorgiou et al., 1996). The effect
of increasing competitor concentration is similar to that of increasing impregnation
time: the catalyst layer is pushed deeper inside the pellet (Hepburn et al., 1989a;
Papageorgiou et al., 1996). For Rh and Pt catalysts prepared using hydrofluoric
acid as competitor, and Pt catalysts prepared using phosphoric acid as competitor,
the local loading of the catalyst layer decreased as it was pushed inwards (Hepburn
et al., 1989a; Schwarz and Heise, 1990). This is because these coimpregnants not
only compete with catalyst precursor for active sites, but also alter the pH of the
pore solution and hence local adsorption capacity of the support.
Another variable in coimpregnation is the nature of the solvent, which however
has been only rarely investigated. For example, Alexiou and Sermon (1993) im-
pregnated silica with CuCl2 and NiCl2 using methanol instead of water as solvent,
and found that the metals penetrated deeper into the support.
Multicomponent coimpregnation can be performed starting with dry or wet
support. The effect on catalyst loading is similar to that discussed earlier for single-
component impregnation: higher total loadings are obtained by dry impregnation
(Hepburn et al., 1991c; Melo et al., 1980b). Hepburn et al. (1991c) observed that
the total loading of egg-white Rh/γ -Al2 O3 catalysts increased with impregnation
time in a manner depending on the initial state of support. For dry impregna-
tion, the rhodium concentration increased mainly towards the external surface
of the pellet, due to continued diffusion of rhodium precursor from the bulk liq-
uid, after imbibition was complete. For wet impregnation, it increased uniformly
in the region between the rhodium layer and the external surface. In addition,
wet impregnation gave sharper catalyst layers than dry impregnation. This was
attributed to faster adsorption and desorption rates relative to the transport rate
for wet impregnation.
As discussed in section 7.2.1, the compounds used in multicomponent im-
pregnation can be introduced simultaneously (coimpregnation) or successively
(sequential impregnation) (Chen et al., 1980; Melo et al., 1980b; Papageorgiou

et al., 1996). Papageorgiou et al. (1996) prepared Pt/γ -Al2 O3 catalysts by impreg-
nating the support with hexachloroplatinic acid and citric acid. When the pellet
was impregnated with catalyst precursor and competitor sequentially, in that or
the reverse order, similar egg-white catalyst distributions were obtained. The main
difference was that if the metal precursor was impregnated first, the catalyst layer
was thinner (Papageorgiou, 1984). Similar egg-white distributions were also ob-
tained with coimpregnation, as discussed later in section 7.2.3. In coimpregnation
experiments, platinum loading of the samples was affected not only by the initial
hexachloroplatinic acid concentration, but also, to a smaller extent, by the citric
acid concentration. The loading decreased with higher initial concentration of cit-
ric acid. However, during sequential impregnation, the platinum loading of the
pellets depended primarily on the conditions of the first impregnation step. Ap-
proximately constant platinum loading was observed for pellets with the same first
impregnation step, regardless of the citric acid concentration in the second step.
According to the classification scheme of Schwarz and Heise (1990) discussed
in section 7.1.2.e, the above coimpregnants belong to the third class, because they
compete with catalyst precursor for adsorption sites. However, other compounds
belonging to the first or second class have also been used to influence the precursor
distribution. As an example of the use of class 2 coimpregnants, Goula et al. (1992a)
prepared eggshell and uniform molybdenum catalysts by heptamolybdate impreg-
nation using acidic and alkaline solutions, respectively. In this case, the amount of
Mo that can be deposited on γ -Al2 O3 increases as the pH is decreased. By succes-
sive impregnation with NH4 OH solution, molybdena species adsorbed towards
the pellet exterior were forced to desorb; hence eggshell and uniform distributions
yielded egg-white and egg-yolk catalysts, respectively. Gaseous competitors have
also been used to affect the precursor profile. Preadsorption of carbon dioxide
on alumina has been found to promote uniform distribution of platinum, iridium,
and rhenium (Kresge et al., 1992).
Multicomponent impregnation is also used for preparing multimetallic catalysts.
Ardiles et al. (1986) and De Miguel et al. (1984) prepared Pt–Re/γ -Al2 O3 cata-
lysts by coimpregnation of H2 PtCl6 and HReO4 . Eggshell catalysts were formed
with layers of Pt and Re largely overlapping, with Pt more concentrated towards
the pellet external surface. This is due to the higher affinity of chloroplatinic
acid for the γ -Al2 O3 support. If HCl was used as a coimpregnant, more uniform
distributions of both metals were obtained. No significant differences were ob-
served between distributions obtained with dry and with wet impregnation. On
the other hand, Summers and Hegedus (1978) prepared Pt–Pd, Lyman et al. (1990)
Pt–Rh, and Hegedus et al. (1979a) Pt–Pd–Rh catalysts, with separated layers of
active materials, by successive impregnations with intermediate drying and cal-
cination. Impregnation of the first metal precursor without competitor produces
an eggshell layer, and subsequent impregnation of the second (and third) metal
precursor along with a competitor deposits it in a subsurface band. When both
metal precursors are impregnated simultaneously, they deposit in the same region
when the support is initially dry, i.e., at the surface if no coimpregnant is used,
and at a subsurface layer if a coimpregnant is present. If the support is prewetted,

a slight separation of the catalyst peaks is observed (Lyman et al., 1990). Cheng
and Pereira (1987) prepared cobalt–molybdenum catalysts by incipient wetness
impregnation and found that the molybdenum profile was of the eggshell type,
while the cobalt was distributed more uniformly. The penetration depth of molyb-
denum increased with decreasing surface area of the support. Thus an interesting
effect of the support surface area, which determines the number of active sites
available for adsorption, was demonstrated.
7.2.2.c Effects of Drying

The drying step which follows the impregnation phase causes evaporation and
flow of liquid in the pores, which can result in redistribution of impregnant in the
pore volume. This can influence the adsorption–desorption equilibrium, leading
to additional precursor adsorption, or affect solute precipitation from the pore-
filling solution (Kheifets et al., 1979). The latter is the dominant mechanism of
catalyst formation for weakly adsorbing precursors, but it can also be important
for strongly adsorbing precursors if the initial concentration of the impregnating
solution is greater than that required to saturate the support.
Drying of porous materials follows various stages characterized by drying rate
(constant or falling), liquid-phase distribution (continuous or pendular), and pre-
dominant mechanism of moisture transport (capillary flow or vapor diffusion)
(Berger and Pei, 1973; Neimark et al., 1981; Lee and Aris, 1985; Coulson and
Richardson, 1991; Moyers and Baldwin, 1997). Two limiting regimes have been
distinguished (see Figure 7.11): fast drying where vapor removal is much faster
than capillary flow (Jv Jc ), and slow drying in the opposite case. In fast drying
Figure 7.11. Regimes of drying. I, Fast drying.

II, Slow drying: (a)–(d) indicate the various
steps.
the evaporation front is continuous and recedes towards the pellet center as dry-
ing progresses, while in slow drying the evaporation front is initially located at
the pellet surface, later portions of it move inward, and finally it breaks down
so that isolated domains of liquid phase are formed (Neimark et al., 1976, 1981;
Fenelonov et al., 1979). Evaporation at the pellet surface and at a moving front
receding towards the pellet interior, depending on the drying conditions, have
been experimentally observed (Hollewand and Gladden, 1992; Gladden, 1995).
In addition, it has been shown that intrapellet moisture distribution can be quite
different for pellets with different pore size distribution even when the overall
drying rate is the same (Hollewand and Gladden, 1994).
The redistribution of solute is affected not only by liquid capillary flow but also
by diffusive transport of the solute. The nonvolatile solute concentration increases
in regions where the liquid vaporizes, thus creating concentration gradients, which
cause diffusion of the solute (Lee and Aris, 1985). If it is fast enough as compared
to vapor removal, diffusional relaxation smooths out concentration gradients. In
the opposite case, high solute concentration exists at the evaporation front, and
if it exceeds the saturation concentration, crystallites are formed. The interac-
tion between mass transport and crystallite nucleation/growth can affect solute
concentration gradients and catalyst dispersion, as discussed in section 2.4.1.
From the above discussion it follows that in general the solute distribution is
determined by the relative magnitudes of evaporation rate, capillary flow, and so-
lute diffusion. Since capillary flow is much faster than diffusive transport, for the
purposes of precursor redistribution it is more appropriate to define as fast drying
the regime where liquid evaporation is much faster than capillary flow (Jv Jc ),
as slow drying the regime where liquid evaporation is much slower than diffusive
transport (Jd Jv ), and as intermediate drying the regime where liquid evapo-
ration is much slower than capillary flow but much faster than diffusive transport
(Jc Jv Jd ). With respect to the final precursor concentration profile in the
pellet, fast drying allows one to preserve the same profile present in the pore-filling
liquid at the end of the impregnation step, i.e., no redistribution of the precursor
occurs. In all other cases the precursor concentration profile in the pore liquid
changes during drying and consequently alters the deposited precursor profile
in the pellet. The final profile is the result of complex interactions among the
transport processes mentioned above, and is also affected by adsorption and pre-
cipitation in addition to the local and global pore size distribution and connectivity.
Other processes such as solvent adsorption, film flow (i.e. two-phase capil-
lary flow) and recondensation through the gas phase may affect liquid transfer
(Fenelonov, 1975; Neimark et al., 1981). Because of the large number of param-
eters and their interactions, it is difficult to draw conclusions of general valid-
ity about the effects of drying on catalyst distribution. Accordingly, conclusions
reported in the literature, particularly in experimental work, are limited to the
specific systems and conditions investigated. In the following we report some of
these, to illustrate the effects of drying on catalyst distribution, although, owing to
lack of quantitative data about diffusion and capillary flow, they do not provide a
comprehensive picture.
For catalysts produced by nonadsorbable (or weakly adsorbed) impregnants,

drying in the intermediate regime leads to an increase in the concentration of
precursor near the external surface of the pellet, while in the slow regime diffu-
sional relaxation smooths the concentration gradients. Van den Berg and Rijnten
(1979) observed that CuCl2 surface accumulation on γ -Al2 O3 was smaller when
dried slowly. For γ -Al2 O3 pellets impregnated with ammonium heptamolybdate,
the mildest drying conditions resulted in catalysts with more uniform Mo dis-
tribution and higher dispersion (Ochoa et al., 1979; Galiasso et al., 1983; Goula
et al., 1992b). Harriott (1969) obtained a nearly uniform distribution of AgNO3
with a low-surface-area alumina by drying slowly at 40◦ C. However, similar drying
conditions yielded wide eggshell distribution for spheres soaked in silver lactate
solution, indicating transition to the intermediate regime. Komiyama et al. (1980)
found that NiCl2 surface segregation can be suppressed by reaching the drying
temperature of 110◦ C at 600◦ C/h, presumably owing to operation in the fast dry-
ing regime, whereas segregation occurred when drying at 100◦ C/h. On the other
hand, for a similar system, Uemura et al. (1986, 1987) observed nickel segrega-
tion towards the outer surface of the support for high-severity drying conditions
only.
In the case of strongly adsorbing precursors, single-component impregnation
leads to eggshell distributions. Lee and Aris (1985) observed that these were
similar regardless of the rate of drying. Chen et al. (1980) noticed that widen-
ing and smoothing of the eggshell layer of Cr/γ -Al2 O3 and Cu/γ -Al2 O3 catalysts
occurred if they were left standing in air before drying at 150◦ C, and the effect
was more pronounced for less strongly adsorbed solutes. Li et al. (1994) prepared
Ni/Al2 O3 catalysts, and even though uniform profiles were expected because of
the prolonged wet impregnation, eggshell distributions were obtained due to redis-
tribution during drying at 200◦ C. Santhanam et al. (1994) investigated the effect
of adsorption strength by considering several precursors of Pt and Pd with dif-
ferent supports and impregnating conditions. They demonstrated that no solute
migration occurred during drying of an initially homogeneously distributed pre-
cursor when strong interaction between solute and support existed, while surface
segregation occurred for weaker interaction.
In all the above studies, the drying regime was influenced by adjusting the
rate of liquid evaporation, for example through the drying temperature. Another
option is to modify the capillary flow, for example by changing the viscosity of the
impregnating solution. In particular, if the capillary flow of solvent is retarded,
the solvent may not be supplied to the pellet surface at a fast enough rate to
balance the evaporation rate. The liquid–air interface will then retreat towards
the pellet center, until there is enough diffusional resistance between the interface
and the pellet surface so that the evaporation rate is lowered to equal the capillary
flow rate. Kotter and Riekert (1979, 1983b) investigated the effect of increasing
the viscosity of the impregnating solution (by adding hydroxyethylcellulose) on
the distribution of CuO on α-Al2 O3 . An eggshell profile was obtained, which
extended progressively towards the pellet center with increasing viscosity, yielding
ultimately a uniform distribution.
Figure 7.12. Experimentally obtained

rhodium distribution across γ -Al2 O3
slabs coimpregnated with 5.7 × 10−2
M RhCl3 and 6.64 M HF, and dried at
temperatures shown. (From Hepburn
et al., 1989b.)
The effect of drying has also been studied in cases where at the end of impreg-
nation an egg-white catalyst distribution is obtained. As expected, the active layer
is preserved by fast drying, since the adsorbed precursor does not have time to
redistribute (Hepburn et al., 1989b). Milder drying conditions can cause blurring
of the catalyst layer, which becomes diffuse and wider and moves inwards (Lee
and Aris, 1985). In addition, Hepburn et al. (1989b) and Becker and Nuttall (1979)
noticed that in such cases, catalyst became concentrated also towards the external
surface, giving rise to bimodal distributions, as shown in Figure 7.12. By drying
slowly at 20◦ C, all peaks were completely washed out, giving a uniform catalyst
(Hepburn et al., 1989b). In light of the discussion earlier in this section, it appears
that drying at 20◦ C corresponds to slow, at 300 and 400◦ C to intermediate, and at
500◦ C to fast drying. Finally, two other alternatives for minimizing catalyst redis-
tribution during drying are worth mentioning: immobilization of precursor before
drying through liquid-phase reduction (Harriott, 1969; Becker and Nuttall, 1979;
Kunimori et al., 1982), and freeze drying with liquid nitrogen (Chu et al., 1989).
7.2.2.d Determination of Catalyst Distribution

Measurements of local catalyst loading are required to establish the catalyst dis-
tribution. This is done by sectioning the catalyst pellet and then analyzing the
internal catalyst concentration profile with a suitable technique. Although we do
not describe these techniques, it is useful to list those that have been employed in
the experimental studies reported in the literature. The most common technique
is electron probe microanalysis (EPMA); other methods include light transmis-
sion, staining, spectroscopic analysis, and autoradiography. A summary of catalyst
systems investigated using these methods is given in Table 7.3.
It is worth noting that measurement of the catalyst distribution, i.e. local load-
ing as a function of position, does not necessarily mean that the behavior of the
pellet under reaction conditions can be predicted, even for structure-insensitive
reactions. The reason is that the reaction rate is proportional to the number of
active sites, which is proportional to the catalytic surface area and not to the
loading. The number and type of active sites are generally influenced by the con-
ditions employed during impregnation. For example, it has been observed that
adsorption can lead to species more difficult to reduce than those formed by pre-
cipitation from the pore-filling solution (Baiker and Holstein, 1983; Baiker et al.,
1986; Goula et al., 1992a).
7.2.3 Comparison of Model Calculations with Experimental Studies

In order to improve our understanding of impregnation processes and to deter-
mine the reliability of mathematical models described in the previous section, it
is useful to compare model predictions with experimental results. In the follow-
ing, we discuss several comparisons that have been reported in the literature for
different systems.
7.2.3.a Dry Impregnation

Komiyama et al. (1980) prepared nickel catalysts by dry impregnation of γ -Al2 O3
with NiCl2 . Uniform and eggshell distributions were obtained by changing the
initial concentration of the impregnating solution. Coimpregnation experiments
utilizing HNO3 resulted in egg-yolk distributions. This behavior was rationalized in
terms of a pH effect. Since NiCl2 does not adsorb below pH = 4.5, coimpregnation
with HNO3 inhibits Ni adsorption in the low-pH regions which are encountered
towards the pellet surface. The experiments were simulated with the single-pore
model (7.41), (7.42), which describes dry impregnation during the imbibition pe-
riod. The precursor flux Wi from the impregnating solution to the pore wall was
calculated assuming either mass transfer or adsorption kinetic control. Adsorption
of NiCl2 and HNO3 was described by Langmuir and linear kinetics, respectively.
Table 7.3. Measurement methods of catalyst distribution and systems investigated.
Catalyst System References

A. Electron Probe Microanalysis
Co/Al2 O3 Maitra et al., 1986; Otsuka et al., 1987
Co–Mo/Al2 O3 Cheng and Pereira, 1987
Cr/Al2 O3 Mathur et al., 1972; Chen and Anderson, 1973, 1976;
Chen et al., 1980
Cu/Al2 O3 Chen and Anderson, 1976; Chen et al., 1980;
Van den Berg and Rijnten, 1979; Ott and Baiker, 1983;
Baiker and Holstein, 1983; Baiker et al., 1986
Ir/Al2 O3 Kresge et al., 1992
Mo/Al2 O3 Ochoa et al., 1979; Galiasso et al., 1983;
Duncombe and Weller, 1985; Goula et al., 1992a;
Santhanam et al., 1994
Mo/SiO2 Santhanam et al., 1994
Ni/Al2 O3 Cervello et al., 1976; Komiyama et al., 1980;
Maitra et al., 1986; Fujiyama et al., 1987;
Chu et al., 1989; Uemura et al., 1989;
Li et al., 1994
Ni–Ba/Al2 O3 Melo et al., 1980b
Ni–Mo/Al2 O3 Souza et al., 1989; Gazimzyanov et al., 1992
Ni–P/Al2 O3 Ko and Chou, 1994
Pd/Al2 O3 Satterfield et al., 1969; Hepburn et al., 1989a;
Lin and Chou, 1994, 1995
Pt/Al2 O3 Roth and Reichard, 1972; Becker and Nuttall, 1979;
Van den Berg and Rijnten, 1979; Shyr and Ernst, 1980;
Hepburn et al., 1989a; Kresge et al., 1992
Pt–Bi/C Kimura, 1993
Pt–Rh/Al2 O3 Lyman et al., 1990
Re/Al2 O3 Kresge et al., 1992
Rh/Al2 O3 Hepburn et al., 1989a,b,c, 1991a,b,c
Ru/Al2 O3 Kempling and Anderson, 1970
W/Al2 O3 Maitra et al., 1986
B. Light Transmission
Ni/Al2 O3 Komiyama et al., 1980
Pt/Al2 O3 Heise and Schwarz, 1985, 1986, 1988;
Schwarz and Heise, 1990
Pt–Re/Al2 O3 Ardiles et al., 1986
C. Staining
Mo/Al2 O3 Srinivasan et al., 1979
Pt/Al2 O3 Castro et al., 1983; Papageorgiou et al., 1996
D. Spectroscopic Analysis
Mo/Al2 O3 Blanco et al., 1987; Goula et al., 1992a,b
E. Autoradiography
Ag/Al2 O3 Harriott, 1969
Only the imbibition period was considered. The convective flow rate was evaluated
using the Washburn equation, and diffusive transport was neglected. By fitting
experimental catalyst profiles, they concluded that adsorption was controlled by
kinetics and not by mass transfer. The model was consistent with observed nickel
profiles, for both one- and two-component systems.
Heise and Schwarz (1988) prepared platinum catalysts by dry impregnation of
γ -Al2 O3 pellets with hexachloroplatinic acid. Adsorption equilibrium data were
obtained independently and shown to follow a Langmuir isotherm. In this case
also, the impregnation model (7.41), (7.42) was used, assuming adsorption kinetics
to control the precursor flux from the impregnating solution to the pore wall. The
adsorption kinetic parameters were evaluated by fitting platinum uptake during
the impregnation experiments. The calculated platinum concentration profiles
in the pellets were shown to be in qualitative agreement with the experimental
eggshell profiles.
Platinum profiles obtained after coimpregnation of γ -Al2 O3 with hexachloro-
platinic and phosphoric acids were also simulated (Schwarz and Heise, 1990). For
this, a multicomponent single-pore model for dry impregnation as discussed in
section 7.2.1.a was used (Kulkarni et al., 1981). Phosphoric acid adsorption was
assumed to follow Langmuir kinetics with an adsorption constant three times
greater than that for the platinate ion. Qualitative agreement was obtained only
after effects of the coimpregnant on solution-interface chemistry were incorpo-
rated into the model, viz., the influence of decreasing pH was simulated by de-
creasing the adsorption capacity of the support.
Hepburn et al. (1991a) impregnated dry γ -Al2 O3 honeycombs with rhodium
trichloride and hydrofluoric acid, and obtained egg-white Rh catalyst distributions.
For both impregnants, single-component equilibrium adsorption data were mea-
sured and fitted with Langmuir adsorption isotherms. Multicomponent adsorption
was described by the competitive Langmuir model [see equation (7.13)]. The cata-
lyst preparation was simulated by considering separately the two steps: imbibition
and drying. For the first, the single-pore model (7.41), (7.42) was used, assuming ad-
sorption kinetic control; the second was simulated through a diffusion–adsorption
model, similar to equation (7.44), with appropriate BCs to take account of sol-
vent evaporation. Two adjustable parameters were used, viz. the adsorption rate
constant of RhCl3 and the effective diffusivity of HF. The two models were used
sequentially to reproduce the Rh profile measured experimentally. Good quanti-
tative agreement was obtained by tuning the two adjustable parameters, as shown
in Figure 7.13. The prediction of the peak height was particularly sensitive to
the chosen value of the RhCl3 adsorption rate constant, while the peak location
was primarily controlled by the surface saturation constant determined from the
adsorption equilibrium data.
7.2.3.b Wet Impregnation

For one-component wet impregnation, the diffusion–adsorption model (7.44),
(7.45) is generally used. Under the assumption that precursor adsorption is rapid
Figure 7.13. Comparison of model prediction with ex-

perimental results for Rh distribution across the wall
of γ -Al2 O3 monoliths prepared by coimpregnation with
3.8 × 10−5 -mol/cm3 RhCl3 and 3.3 × 10−3 -mol/cm3 HF.
(From Hepburn et al., 1991a.)
and irreversible, so that the time scale for adsorption is much smaller than that for
diffusion, the model reduces to a shell progressive model. This has been used to
simulate experimental results of CuCl2 (Baiker and Holstein, 1983) and H2 PtCl6
impregnation of γ -Al2 O3 (Santacesaria et al., 1977c). In both cases, effective dif-
fusivities were estimated by fitting the catalyst penetration depth as a function of
time. Baiker and coworkers (Baiker and Holstein, 1983; Baiker et al., 1986) took
into account both adsorbed copper and copper precipitated from the pore solution
during drying, which ultimately gave rise to two types of copper species with dif-
ferent properties as identified by TPR. The agreement between experimental and
calculated catalyst profiles was good, even though the model predicted a slower
approach of the penetration front to the center of the pellet.
Hanika and coworkers (Hanika et al., 1982, 1983; Machek et al., 1983a) pre-
pared Pt catalysts on activated carbon and alumina, utilizing aqueous and or-
ganic solutions of hexachloroplatinic acid, as well as phosphene and amine com-
plexes of platinum. The one-component wet-impregnation model (7.44), (7.45)
was used, assuming local adsorption equilibrium according to Langmuir or Fre-
undlich isotherms. The effective diffusivities were calculated by fitting the time-
dependent Pt uptake from the impregnating solution. The computed catalyst dis-
tributions were in qualitative agreement with the experimental data.
A two-component (RhCl3 –HF) wet-impregnation system was considered by
Hegedus et al. (1979b) using the adsorption–diffusion model (7.44). The multi-
component adsorption was described by the competitive Langmuir model (7.13),
whose parameters were estimated from single-component data of adsorption ki-
netics on powder support. In the Langmuir model, the various species were al-
lowed to have different saturation concentrations. Diffusivities were determined
by comparing the measurements of species uptake on pellets, using results of the
diffusion–adsorption model. The model predicted Rh peak locations inside the
γ -Al2 O3 pellets satisfactorily.
A similar approach was followed by Scelza et al. (1986) to simulate wet coim-
pregnation of γ -Al2 O3 pellets with H2 PtCl6 and HCl. The same competitive
Langmuir model was used to describe adsorption kinetics, but now assuming
the same saturation concentration for all components. Moreover, the diffusion
coefficients and the adsorption rate constants were estimated simultaneously by
fitting experimental data on species uptake from the impregnating solution as
a function of time. Unfortunately, different values of the adjustable parameters
were obtained at each acid concentration, although the scatter was significant
only for desorption rate constants. Modeling for various acid concentrations was
carried out utilizing the corresponding set of diffusivities and adsorption param-
eters, and good agreement between theoretical and experimental profiles was
obtained. It was observed that with increasing HCl concentration the profile de-
velops from eggshell to uniform. The same approach was also used to simulate
bimetallic wet coimpregnation. In particular, experimental data on H2 PtCl6 and
HReO4 coimpregnation in the presence of HCl were considered (De Miguel et al.,
1984; Ardiles et al., 1986). The predicted and experimental profiles of the various
species after 12-h impregnation were in good agreement except the Re profiles,
for which the predicted trend to form an egg-white distribution was not confirmed
experimentally. Finally, it was found that the model was also able to predict Pt,
Re, and Cl profiles obtained by dry impregnation, indicating that the imbibition
period had no important influence due to the long impregnation time used in the
experiments.
Melo et al. (1980b) prepared Ni and Ba catalysts on γ -Al2 O3 by wet and dry
impregnation. They performed single- and two-component experiments, where
Ni(NO3 )2 and Ba(NO3 )2 were impregnated simultaneously or successively. In
the latter case, Ni was impregnated first, and then, after calcination, Ba was im-
pregnated as the second component. Wet impregnation was simulated with the
adsorption–diffusion model (7.44), introducing external resistance to mass trans-
fer and assuming local adsorption equilibrium inside the pellet. This was described
through the multicomponent Langmuir model, whose parameters were estimated
from independent equilibrium data (Melo et al., 1980a). The results of the wet-
impregnation model were used as initial conditions for the drying model, to take
proper account of the effect of drying on catalyst distribution. For dry impregna-
tion a simplified model was used to describe the first imbibition period, and the
wet-impregnation model was used again for the subsequent part of the process.
The simplified model was based on the assumption that the penetration rate is
much higher than the adsorption rate, so that adsorption could be neglected dur-
ing imbibition (Melo et al., 1980b). From the first set of experiments involving
one-species impregnation, it was found that diffusion and external mass trans-
fer coefficients could not be evaluated independently but had to be estimated by
fitting the species uptake data as a function of time. On the other hand, these
parameter values allowed them to predict with reasonable accuracy the catalyst
profiles obtained in two-species impregnation experiments. It is worth noting that
to improve agreement with the experimental data, an adjustable parameter was
introduced. This was the fraction of adsorption sites postulated to be accessible
to nickel, while the remaining ones were available for both metals. On the whole,
experimental profiles for one-component impregnation exhibited a weak eggshell
shape, while the calculated ones were sharper. Two-component impregnations
yielded more uniform distributions.
In the above studies, effective diffusivities were estimated by comparing ex-
perimental solute uptake measurements in pellets with results of a diffusion–
adsorption model. In some cases, not only diffusivities but also other parameters
(e.g. adsorption constants, external mass transfer coefficients) were adjusted to
fit experimental data. Price and Varma (1988) developed an experimental proce-
dure for the independent evaluation of diffusivity, based on the idea of decou-
pling adsorption and diffusion. This was done by running diffusion experiments
at sufficiently high solute concentrations. In Langmuir adsorption, surface con-
centration of adsorbate does not change appreciably at saturation conditions, i.e.
for fluid concentration values beyond a sufficiently high value. Thus, on keeping
both the pore and the bulk fluid in this concentration range, adsorption and des-
orption do not occur. Hence, the diffusion–adsorption model equations (7.44),
(7.45) are simplified, in that, ∂n/∂t = 0. In order to ensure saturation conditions,
pellets were placed in a high-concentration solution for several days, prior to
their use in diffusivity experiments. They were then moved to a solution of dif-
ferent (but still sufficiently high) concentration. Effective diffusivities were deter-
mined by solving equations (7.44), (7.45) and calculating the value of De which
gave the best fit to the time-dependent solute concentration change in the bulk
solution.
Papageorgiou et al. (1996) prepared Pt/γ -Al2 O3 pellets with a sharp internal
step distribution of catalyst using coimpregnation and sequential impregnation of
hexachloroplatinic and citric acids. In the sequential impregnation, the catalyst
precursor was impregnated first, and then the competitive adsorbate. The latter
adsorbed more strongly and pushed the former towards the center of the pellet,
resulting in egg-white catalysts. Impregnation in the reverse order also produced
similar egg-white catalysts, because the catalyst precursor moved past the pellet
portion where citric acid had adsorbed (Papageorgiou, 1984). Similar behavior was
observed for γ -Al2 O3 pellets impregnated in a sequential fashion with chromium
nitrate (strongly adsorbed) and copper nitrate (less strongly adsorbed) (Chen
et al., 1980).
The cross sections of pellets prepared by sequential and coimpregnation tech-
niques are shown in Figure 7.14. The dark rings in Figure 7.14.a–i are the platinum
catalyst, while the lighter portion is the γ -Al2 O3 support. For comparison, eggshell,
egg-yolk, and uniform pellets are also shown in Figure 7.14.j–l. For both types of
impregnations, the platinum band was located deeper inside the pellet for higher
initial citric acid concentration (compare Figure 7.14.b–c, e–f, h–i). In addition,
longer impregnation time forced the platinum to move closer to the pellet cen-
ter, since the citric acid could diffuse further inside the pellet and displace the
adsorbed platinum (compare Figure 7.14.a–b, d–e, g–h).
Increasing the initial hexachloroplatinic acid concentration did not have any
noticeable effect on catalyst location; however, it resulted in catalysts with higher
Figure 7.14. Step-type platinum catalysts prepared by sequential impregnation:

(a) 0.00589 M hexachloroplatinic acid (15 min), 0.1111 M citric acid (15 min),
(b) 0.00589 M hexachloroplatinic acid (15 min), 0.1111 M citric acid (60 min),
(c) 0.00589 M hexachloroplatinic acid (15 min), 0.3333 M citric acid (60 min), (d)
0.0117 M hexachloroplatinic acid (15 min), 0.1111 M citric acid (15 min), (e) 0.0117
M hexachloroplatinic acid (15 min), 0.1111 M citric acid (60 min), (f) 0.0117 M
hexachloroplatinic acid (15 min), 0.3333 M citric acid (60 min); and by coimpreg-
nation: (g) 0.00589 M hexachloroplatinic acid and 0.1111 M citric acid (15 min), (h)
0.00589 M hexachloroplatinic acid and 0.1111 M citric acid (60 min), (i) 0.00589 M
hexachloroplatinic acid and 0.1667 M citric acid (60 min), (j) 0.0117 M hexachloro-
platinic acid (15 min), (k) 0.0117 M hexachloroplatinic acid and 0.1667 M citric acid
(3 h), (l) 0.0117 M hexachloroplatinic acid and 0.1667 M citric acid (10 h). (From
loading and somewhat larger bandwidths (compare Figure 7.14.a–d, b–e, c–f).
The bandwidth was also influenced by the impregnation time, owing primarily to
geometric effects. Since the diameter of the ring where the platinum was located
decreased with time, its width increased to accommodate the deposited metal
(compare Figure 7.14.a–b, d–e, g–h). In general, for the same impregnation time
and adsorbate concentrations, similar penetration depths were achieved by the
Figure 7.15. Comparison of model

predictions and experimental mea-
surements of the surface concen-
tration profiles of platinum for se-
quential impregnation with 0.1111 M
citric acid and (a) 0.00589 M hex-
achloroplatinic acid, (b) 0.0117 M
hexachloroplatinic acid. The shaded
bands indicate the experimentally
measured platinum step positions,
and the numbers on the bands give
the impregnation time in the citric
acid solution. (From Papageorgiou
et al., 1996.)
two types of impregnations (compare Figure 7.14.a–g, b–h), but the bandwidth
was smaller (as small as 50 µm) when using sequential impregnation.
The eggshell distribution (Figure 7.14.j) was prepared by impregnating the pel-
let in a solution of hexachloroplatinic acid alone. The egg-yolk distribution (Figure
7.14.k) could be obtained by both sequential impregnation and coimpregnation,
provided that the impregnation periods were sufficiently long. The uniform distri-
bution (Figure 7.14.l) was prepared using the coimpregnation method. Note that
the difference in the preparation of pellets depicted in Figure 7.14.k and l is only in
the impregnation duration. Thus, for short coimpregnation time an egg-white de-
position is obtained, while for longer impregnation times egg-yolk and ultimately
uniform distributions are attained.
Modeling of coimpregnation and sequential impregnation of pellets was car-
ried out using equations (7.44), (7.45). Figures 7.15 and 7.16 show the predicted
dimensionless platinum surface concentration profiles, along with the experimen-
tally measured platinum band positions denoted by the shadowed regions for the
sequential-impregnation and coimpregnation studies, respectively. The numbers
on the bands give the impregnation time (for sequential impregnation it is the
time in citric acid solution). The penetration depth of platinum predicted by the
Figure 7.16. Comparison of model

predictions and experimental mea-
surements of the surface concen-
tration profiles of platinum for
coimpregnation with 0.0117 M hex-
achloroplatinic acid and (a) 0.0556 M
citric acid, (b) 0.1111 M citric acid.
The shaded bands indicate the experi-
mentally measured platinum step po-
sitions, and the numbers on the bands
give the impregnation time. (From
model for both impregnation techniques, increases with increasing impregnation

time, as also observed experimentally. In addition, the model is able to predict that
the bandwidths are broader for coimpregnation than for sequential impregnation
(compare Figures 7.15.b and 7.16.b), as well as the fact that the bandwidth increases
with impregnation time. Note the model predicts that some adsorbed platinum
extends from the peak to the pellet surface, with coimpregnation exhibiting more
platinum in that region than sequential impregnation. This is consistent with ex-
perimental observations by various investigators (Kunimori et al., 1982; Schwarz
and Heise, 1990; Hepburn et al., 1989a).
Figure 7.17 compares model predictions for the platinum peak location with
experimental data, for various citric acid concentrations and impregnation times.
The model agrees well with the data and demonstrates that on increasing the citric
acid concentration for both impregnation techniques, the platinum band moves
deeper within the pellet. The maximum difference between the experimental and
theoretical peak locations is less than 8% of the pellet radius.
Chu et al. (1989) used the triple-layer electrostatic model discussed in sec-
tion 7.1.3.d to simulate adsorption of NiCl2 in the presence of HCl and NaCl
on wet γ -Al2 O3 pellets. As compared to Langmuir-based models, this provides a
more realistic picture of surface chemistry, at the expense of model complexity.
Figure 7.17. Experimental platinum peak lo-

cations and model predictions using 0.00589
M hexachloroplatinic acid and various cit-
ric acid concentrations: (a) sequential im-
pregnation, and (b) coimpregnation. (From
Adsorption of ions was described by the following reactions:

+
& S–O− NiOH + 2H+
S–OH + H2 O + Ni2+ % (7.48)
− −
+
S–OH + H + Cl %
& S–OH+
2 Cl (7.49)
S–OH + Na+ % +
& S–O− Na + H+ . (7.50)
The support surface was assumed to be in equilibrium with the local pore concen-
trations of HCl and NiCl2 . The triple-layer model resulted in a system of twelve
equations comprising charge–potential relationships, the electroneutrality condi-
tion, material balance for adsorption sites, charge balance equations, and mass
action equations. The diffusion–adsorption equations (7.44) were used, but only
for diffusion of H+ , OH− , and Ni2+ , while the Na+ and Cl− pore concentrations
were assumed to be constant throughout the pellet. The various model parameters
were determined independently from adsorption measurements, and the effective
diffusion coefficients were calculated from literature values for bulk diffusivities
and a tortuosity factor of 2. Only the effective diffusivity of hydrogen ions was
adjusted to obtain agreement with experimental results. An important feature of
this model is the generation of protons upon nickel adsorption as described by
equation (7.48), which explains the origin of the depression of Ni uptake under
acidic conditions, also observed by Komiyama et al. (1980) for a NiCl2 /HNO3
system as discussed in section 7.2.3.a. As nickel diffuses inside the pellet and ad-
sorbs, a local excess of protons is generated, reducing nickel uptake and giving
rise to a pH profile inside the pellet. This produces unique concave–convex metal
profiles in the pellets, which were verified experimentally. These are maintained
Figure 7.18. Nickel profile in a γ -alumina pellet after 26.75-h

impregnation with 10−3 M NiCl2 and 0.1 M NaCl, after acid
pretreatment for 120 h. (From Chu et al., 1989.)
even after long impregnation times, due to the slow migration of protons rela-
tive to the metal ions. By controlling the initial pH profile, eggshell, egg-white,
egg-yolk, and uniform catalysts can be obtained. For example, for producing egg-
white catalysts, an acid pretreatment is required to yield a region of lower pH at
the pellet surface than in the interior, followed by nickel impregnation. Nickel
ions diffuse relatively quickly past the lower pH shell and adsorb as they reach the
higher-pH internal region, creating an egg-white profile. For longer impregnation
times, nickel diffusion continues and the egg-white region enlarges to produce an
egg-yolk catalyst. Agreement of the model calculations with experimental nickel
distributions was good in most cases, as shown in Figure 7.18.
APPENDIX A
Application of the Maximum Principle for

Optimization of a Catalyst Distribution
The optimal catalyst distributions in a pellet and in an inert membrane reactor

reported in sections 2.5.1 and Appendix B, respectively, have been obtained using
the maximum principle. In this appendix we report the details of that analysis.
In both cases three problems are addressed, which differ only in the objective
function. They can all be reduced to the same form: determine the control function
µ(s) which, under the constraints
dx1
= f1 (x2 ) (A.1)
ds
dx2
= f2 (x1 , µ) (A.2)
ds
dx3
= f3 (µ) (A.3)
ds
dx4
= f4 (x1 , µ) (A.4)
ds
dx5
= f5 (x1 , µ) (A.5)
ds
with
dx1
x2 = s n (A.6)
ds
and BCs
x2 = 0 at s = 0 (A.7)
x2 = A [f(x2 ) − x1 ] at s = 1 (A.8)
x3 = 0 at s = 0 (A.9)
1
x3 = at s = 1 (A.10)
n+1
x4 = 0 at s=0 (A.11)
x5 = 0 at s=0 (A.12)
maximizes the functional
1
I = G [x2 |1 , x4 |1 , x5 |1 ] + R(x1 , µ) ds. (A.13)
0
181
182 Application of the Maximum Principle for Optimization
These equations represent both the pellet and the membrane problems through
the following assignments:
Pellet problem.
Objective Effectiveness factor Selectivity Yield
function η S Y
x1 x x x
x2 γ γ γ
x3 h h h
x4 0 xN 0
x5 0 xM 0
f1 γ s −n γ s −n γ s −n
µ(s) µ(s) µ(s)
f2 w(x)s n w(x)s n w(x)s n
1 + Bµ(s) 1 + Bµ(s) 1 + Bµ(s)
f3 µ(s) s n µ(s) s n µ(s) s n
µ(s)
f4 0 w N (x)s n 0
1 + Bµ(s)
µ(s)
f5 0 w M (x)s n 0
1 + Bµ(s)
f xf xf xf
I ηM SNM YNM
DM xN (1)
G 0 0
DN xM (1)
µ(s) µ(s)
R w M (x)s n 0 w N (x)s n
1 + Bµ(s) 1 + Bµ(s)
Membrane problem.
Objective Conversion Selectivity Purity
function X S P
x1 x x x
x2 γ γ γ
x3 h h h
x4 0 0 0
x5 0 0 0
f1 γ s −n γ s −n γ s −n
µ(s) µ(s) µ(s)
f2 w(x)s n w(x)s n w(x)s n
1 + Bµ(s) 1 + Bµ(s) 1 + Bµ(s)
f3 µ(s) s n µ(s) s n µ(s) s n
f4 0 0 0
f5 0 0 0
f xf = p3 [γ(1)] xf = p3 [γ(1)] xf = p3 [γ(1)]
I XN SDN PDN
G (B.27) (B.30) (B.32)
R 0 0 0
Application of the Maximum Principle for Optimization 183
The optimization problem can be solved using the variational method described
in more general terms in Ray (1981). Here we show its application to the cases of
interest. Let us assume that µ̄(s) is the optimal control function and express any
other control function infinitesimally close to it as a perturbation about µ̄(s):
µ(s) = µ̄(s) + δµ(s). (A.14)
The state x(s) resulting from µ(s) can also be represented as a perturbation about
the state x̄(s) caused by the control function µ̄(s):
x(s) = x̄(s) + δx(s). (A.15)
Next we linearize equations (A.1)–(A.5) about the optimal function µ̄, to yield
d(δx1 ) ∂fT
= 1 δx2 (A.16)
ds ∂x2
d(δx2 ) ∂fT ∂f2
= 2 δx1 + δµ (A.17)
ds ∂x1 ∂µ
d(δx3 ) ∂ f3
= δµ (A.18)
ds ∂µ
d(δx4 ) ∂ f4 ∂ f4
= δx1 + δµ (A.19)
ds ∂x1 ∂µ
d(δx5 ) ∂ f5 ∂ f5
= δx1 + δµ. (A.20)
ds ∂x1 ∂µ
Similarly, equation (A.13) gives

1
∂G ∂G ∂G ∂R ∂R
δI = δx2 |1 + δx4 |1 + δx5 |1 + δx1 + δµ ds. (A.21)
∂x2 ∂ x4 ∂ x5 0 ∂x1 ∂µ
Let us now adjoin to the objective functional the linearized constraints [equations
(A.16)–(A.20)] multiplied by the adjoint variables (Lagrange multipliers) λ i (s):

1
d(δx1 ) ∂ f1T
λT1 (s) − δx2 ds = 0 (A.22)
0 ds ∂x2

1
d(δx2 ) ∂f2T ∂f2
λT2 (s) − δx1 − δµ ds = 0 (A.23)
0 ds ∂x1 ∂µ

1
d(δx3 ) ∂ f3
λ3 (s) − δµ ds = 0 (A.24)
0 ds ∂µ

1
d(δx4 ) ∂ f4 ∂ f4
λ4 (s) − δx1 − δµ ds = 0 (A.25)
0 ds ∂x1 ∂µ
1
d(δx5 ) ∂ f5 ∂ f5
λ5 (s) − δx1 − δµ ds = 0. (A.26)
0 ds ∂x1 ∂µ
If we require that equations (A.16)–(A.20) be satisfied everywhere, then the sub-

traction of equations (A.22)–(A.26) from equation (A.21) yields
∂G ∂G ∂G
δI = δx2 |1 + δx4 |1 + δx5 |1
∂x2 ∂ x4 ∂ x5
1 1
∂R ∂f T
∂ f4 ∂ f5 ∂fT
+ + λT2 2 + λ4 + λ5 δx1 ds + λT1 1 δx2 ds
0 ∂x1 ∂x1 ∂x1 ∂x1 0 ∂x2

1
∂R ∂f2 ∂ f3 ∂ f4 ∂ f5
+ + λT2 + λ3 + λ4 + λ5 δµ ds
0 ∂µ ∂µ ∂µ ∂µ ∂µ
1
d(δx1 ) d(δx2 ) d(δx3 ) d(δx4 ) d(δx5 )
− λT1 + λT2 + λ3 + λ4 + λ5 ds.
0 ds ds ds ds ds
(A.27)
Integrating by parts the various terms of the last integral, we obtain

1 1
T d(δx1 ) dλT1
λ1 ds = λ1 δx1 |1 − λ1 δx1 |0 −
T T
δx1 ds (A.28a)
0 ds 0 ds
1 1 T

d(δx 2 ) dλ
λT2 ds = λT2 δx2 |1 − λT2 δx2 |0 − δx2 2 ds (A.28b)
0 ds 0 ds
1 1
d(δx3 ) dλ3
λ3 ds = λ3 δx3 |1 − λ3 δx3 |0 − δx3 ds (A.28c)
0 ds 0 ds
1 1
d(δx4 ) dλ4
λ4 ds = λ4 δx4 |1 − λ4 δx4 |0 − δx4 ds (A.28d)
0 ds 0 ds
1 1
d(δx5 ) dλ5
λ5 ds = λ5 δx5 |1 − λ5 δx5 |0 − δx5 ds. (A.28e)
0 ds 0 ds
Substituting the above equations (A.28a)–(A.28b) and using the initial conditions
(A.9)–(A.12), the first and last terms of the rhs of equation (A.27) become

∂G 1
d(δx1 ) d(δx2 ) d(δx3 ) d(δx4 ) d(δx5 )
δx2 |1 − λT1 + λT2 + λ3 + λ4 + λ5 ds
∂x2 0 ds ds ds ds ds
∂G
= δx2 |1 − λT1 δx1 |1 + λT1 δx1 |0 − λT2 δx2 |1 + λT2 δx2 |0 − λ4 δx4 |1 − λ5 δx5 |1
∂x2
1
dλT1 dλT2 dλ3 dλ4 dλ5
+ δx1 + δx2 + δx3 + δx4 + δx5 ds. (A.29)
0 ds ds ds ds ds
Now let us examine the terms

∂G
δx2 |1 − λT1 δx1 |1 + λT1 δx1 |0 − λT2 δx2 |1 + λT2 δx2 |0 . (A.30)
∂x2
From the initial conditions (A.7) and (A.8), we see that
δx2 |0 = 0 (A.31)
and

δx2 |1 = A J1 δx2 |1 − δx1 |1 (A.32)
where J1 is the Jacobian of f with respect to x2 . Solving for δx2 |1 , we get

I − A J1 δx2 |1 = −A δx1 |1
and then
−1
δx2 |1 = − I − A J1 A δx1 |1 . (A.33)
Substituting (A.31) and (A.33) in (A.30), we have

∂G −1 −1
− I − A J1 A δx1 |1 − λT1 δx1 |1 + λT1 δx1 |0 + λT2 I − A J1 A δx1 |1 .
∂x2
(A.34)
We want this expression to be zero for all δx1 |0 and δx1 |1 ; thus we set
λT1 = 0 at s=0 (A.35)

∂G −1 −1
−λT1 − I − A J1 A + λT2 I − A J1 A = 0 at s = 1. (A.36)
∂x2
Let us now elaborate (A.36):
−1 ∂G −1
λT2 I − A J1 = λT1 A−1 + I − A J1
∂x2
∂G
λT2 = λT1 A−1 I − A J1 +
∂x2
∂G
λT2 = λT1 A−1 − λT1 J1 +
∂x2

T ∂G T
λ2 = A−1 λ1 − JT1 λ1 + .
∂x2
So the BCs for λ1 and λ2 are
λ1 = 0 at s=0 (A.37)
T
T ∂G
λ2 = A−1 λ1 − JT1 λ1 + at s = 1. (A.38)
∂x2
Note that since for the pellet problem neither f nor G is a function of x2 , the BC
(A.38) becomes
T
λ2 = A−1 λ1 at s = 1
which is equivalent to
λT2 = λT1 A−1

λT1 = λT2 A (A.39)
λ1 = AT λ2
while for the membrane problem

∂G T
=v
∂x2
where v is given by equations (B.42)–(B.44). Equation (A.27) now becomes
1 1
∂ H dλT1 ∂ H dλT2
δI = + δx1 ds + + δx2 ds
0 ∂x1 ds 0 ∂x2 ds

1
∂ H dλ3 1
∂ H dλ4
+ + δx3 ds + + δx4 ds
0 ∂ x3 ds 0 ∂ x4 ds

1
∂ H dλ5 1
∂H
+ + δx5 ds + δµ ds
0 ∂ x5 ds 0 ∂µ

∂G ∂G
+ − λ4 δx4 |1 + − λ5 δx5 |1 (A.40)
∂ x4 ∂ x5
where
H = R + λT1 f1 + λT2 f2 + λ3 f3 + λ4 f4 + λ5 f5 (A.41)
is called the Hamiltonian. In order to express the direct influence of δµ alone, let
us choose the arbitrary functions λ such that they satisfy
dλT1 ∂H
=− (A.42)
ds ∂x1
dλT2 ∂H
=− (A.43)
ds ∂x2
dλ3 ∂H
=− (A.44)
ds ∂ x3
dλ4 ∂H
=− (A.45)
ds ∂ x4
dλ5 ∂H
=− . (A.46)
ds ∂ x5
This allows the influence of the system equations (A.1)–(A.5) to be transmitted by
λ and to be felt in ∂H/∂µ. For the terms outside the integral in (A.40) we choose
∂G
λ4 = at s=1 (A.47)
∂ x4
∂G
λ5 = at s = 1. (A.48)
∂ x5
Thus, on using equations (A.42)–(A.48), equation (A.40) becomes
1
∂H
δI = δµ ds. (A.49)
0 ∂µ
Equation (A.49) represents the direct influence of the variation δµ on δ I. The

necessary condition for optimality is that δ I ≤ 0 for all possible small variations
δµ, which implies that
∂H
= 0. (A.50)
∂µ
This condition is used in sections 2.5.1 and 5.2 to identify the optimal catalyst
distribution µ(s) in various cases.
APPENDIX B
Optimal Catalyst Distribution in

Pellets for an Inert Membrane Reactor:
Problem Formulation
In this appendix, equations describing the mass and energy balances in pellets and
in both feed and permeate-side fluid streams of an inert membrane reactor, shown
in Figure 5.1, are presented (cf. Yeung et al., 1994). The problem of determining
the optimal catalyst distribution in pellets is then formulated. Optimization of the
catalyst loading profile is carried out, as in section 2.5.1, using as optimization
variable the dimensionless loading of the active catalyst
q(s)
µ(s) = (B.1)
q¯
which, by definition, satisfies the condition
1
1
µ(s) s n ds = . (B.2)
0 n + 1
On physical grounds, it is assumed that µ(s) is positive and bounded, i.e.
0 ≤ µ(s) < α (B.3)
where α represents the maximum allowable value of the local catalyst loading on
the support.
The dependence of the reaction rate on catalyst loading is represented by equa-
tion (2.129):
µ(s)
r = r0 . (B.4)
1 + Bµ(s)
B.1 The Mass and Energy Balance Equations

The dimensionless steady-state mass and energy balances for the pellet, taking
account of external transport resistances, are given by
J
1 d n dui µ(s)
L[ui ] = n s = Di νi j φ 2j f j (u, θ ) (i = 1, . . . , I)
s ds ds 1 + Bµ(s) j=1
(B.5)
188
Optimal Catalyst Distribution in Pellets for an Inert Membrane Reactor 189
J
1 d n dθ µ(s)
L[θ] = n s =− βj φ 2j fj (u, θ ) (B.6)
s ds ds 1 + Bµ(s) j=1
with BCs
dui
s = 0: =0 (i = 1, . . . , I) (B.7a)
ds
dui
s = 1: = Bim,i (gi − ui ) (i = 1, . . . , I) (B.7b)
ds
dθ
s = 0: =0 (B.7c)
ds
dθ
s = 1: = Bih (τ − θ ) (B.7d)
ds
where the following dimensionless parameters are used:
ui = Ci /C f,1
0
, s = x/R, Di = De,1 /De,i
gi = Cf,i /C f,1
0
, τ = Tf /Tf0

φ 2j = r j Cf,1
0
, Cf,1
0
, . . . , Cf,1
0
, Tf 0 R 2 /De,1 Cf,1
0
0 (B.8)
f j (u, θ) = rj (C1 , C2 , . . . , CI , T)/r j Cf,1 , Cf,1
0
, . . . , Cf,1
0
, Tf 0

θ = T/Tf 0 , β j = (−Hj )De,1 Cf,1 0
/ λe Tf 0
Bim,i = kg,i R/De,i , Bih = hR/λe .
The mass and energy balances can be rewritten in matrix form as follows:
µ(s)
L[x] = w(x) (B.9)
1 + Bµ(s)
with BCs
dx
s = 0: =0 (B.10a)
ds
dx
s = 1: = A(xf − x) (B.10b)
ds
where the vectors x, xf , and w, and the matrix A, are defined by
 
J
 D1 ν1 j φ j f j (x) 
2
 
 j=1 
   J   
u1   g1
 D2 ν2 j φ 2j f j (x) 
 u2     
   j=1   g2 
 ..   .   . 
x =  . , w (x) =  .. ,
 xf =  ..  (B.11)
     
 uI   J   gI 
 DI ν I j φ j f j (x) 
2
θ   τ
 
 j=1

 J 
 − β φ 2 f (x) 
j j j
j=1
190 Optimal Catalyst Distribution in Pellets for an Inert Membrane Reactor
 
Bim,1 0
 
0 Bim,2 0 
 
 
A=

..
.
..
.
..
. .
 (B.12)
 
 0 
 0 Bim,I 
0 Bih
The steady-state mass balances for the cylindrical inert membrane, accounting
for external transport resistances, are given by

1 d dCM,i
DeM,i xM =0 (i = 1, . . . , I) (B.13)
xM dxM dxM
with BCs
dCM,i
xM = xF : −DeM,i = kg,F (CF,i − CM,i ) (B.14a)
dxM
dCM,i
xM = xP : −DeM,i = kg,P (CM,i − CP,i ). (B.14b)
dxM
The mass flux of component i, at the inner or outer membrane surface, can be
obtained on solving the above equations:

dCM,i dCM,i
−2π xP LDeM,i = −2π xF LDeM,i
dxM xM =xP dxM xM =xF
ζF,i PF ζP,i PP
2π LDeM,i Rg TF
− Rg TP
= . (B.15)
ln xP
xF
+ DeM,i kg,F1 xF + kg,P1 xP
Similarly, for the heat flow across the membrane,

dTM dTM
−2π xP LλeM = −2π xF LλeM
dxM xM =xP dxM xM =xF
2π LλeM (TF − TP )
= . (B.16)
ln xP
xF
+ λeM hF1xF + hP1xP
The mass balances for the feed side of the membrane reactor are given by
Vr (1 − ε)(n + 1)De,i CF,1
0
dui (1)
QF0 ζF,i
0
− QF ζF,i −
R2 ds
ζF,i PF ζP,i PP
2π LDeM,i Rg TF
− Rg TP
− 1 =0 (B.17)
ln xP
xF
+ DeM,i kg,F xF + kg,P1 xP
and for the permeate side by

ζF,i PF ζP,i PP
2π LDeM,i Rg TF
− Rg TP
QP0 ζP,i
0
− QP ζP,i + = 0. (B.18)
ln xP
xF
+ DeM,i kg,F1 xF + kg,P1 xP
Additionally, we have the two congruence equations arising from the definition
of the mole fractions in the feed and permeate streams,
CF,i CP,i
ζF,i = I , ζP,i = I (B.19)
i=1 CF,i i=1 CP,i
given by

I
I
ζF,i = 1 and ζP,i = 1. (B.20)
i=1 i=1
The energy balance for the feed side of the membrane reactor is
Vr (1 − ε)(n + 1)λe TF0 dθ(1)

ρcp QF0 TF0
TF0 − (ρcp QF )TF TF −
R2 ds
− =0 (B.21)
ln + λeM hF1xF + hP1xP
xP
xF
and for the permeate side
ρcp QP0 T0
TP0 P
− (ρcp QP )TP TP − = 0. (B.22)
ln + λeM hF1xF + hP1xP
xP
xF
Equations (B.17)–(B.22) represent 2I + 4 linearly independent equations in the

2I + 4 unknowns, ζF,1 , ζF,2 , . . . , ζF,I , ζP,1 , ζP,2 , . . . , ζP,I , QF , QP , TF , and TP . By care-
ful examination, we can see that these equations can be solved once the values
of the derivatives dui (1)/ds and dθ(1)/ds are known. This means that once the
fluxes leaving the catalyst pellets are known, equations (B.17)–(B.22) provide the
values for composition and temperature at the outlet of both feed and permeate
sides. We can formally state this conclusion as follows:
ζ F = p1 [γ(1)] (B.23)
ζ P = p2 [γ(1)] (B.24)
xf = p3 [γ(1)] (B.25)
where p indicates a functional dependence and γ(1) = dx(1)/ds with x defined by

equation (B.11).
B.2 The Performance Indexes

The performance indexes of interest for optimization are:
Conversion of reactant species N:
QF0 ζF,N
0
+ QP0 ζP,N
0
− QF ζF,N − QP ζP,N
XN = (B.26)
QF0 ζF,N
0
+ QP0 ζP,N
0
192 Optimal Catalyst Distribution in Pellets for an Inert Membrane Reactor
which, using equations (B.17) and (B.18), becomes
KN γ N (1)
XN = (B.27)
QF ζF,N + QP0 ζP,N
0 0 0
where
Vr (1 − ε)(n + 1)De,N CF,1
0
KN = − . (B.28)
R2
Selectivity of a product D relative to a reactant N:
QF ζF,D + QP ζP,D
SDN = (B.29)
QF0 ζF,N
0
+ QP0 ζP,N
0
− QF ζF,N − QP ζP,N
which, using a substitution similar to that for conversion, leads to
QF0 ζF,D
0
+ QP0 ζP,D
0
− KDγD(1)
SDN = . (B.30)
KN γN (1)
Purity of a product D relative to a reactant N in the feed side:
ζF,D
PDN = (B.31)
ζF,N
which, using equation (B.23), reduces to
p1D [γ(1)]
PDN = = D[γ(1)]. (B.32)
p1N [γ(1)]
B.3 Development of the Hamiltonian

Equations (B.2), (B.9), (B.10) for the pellet are equivalent to the following set of
first-order differential equations:
dx
= γ s −n (B.33)
ds
dγ µ(s)
= w(x) s n (B.34)
ds 1 + Bµ(s)
dh
= µ(s) s n (B.35)
ds
with BCs
s = 0: γ=0 (B.36a)
s = 1: γ = A(xf − x) (B.36b)
s = 0: h=0 (B.36c)
1
s = 1: h= . (B.36d)
n+1
We can now introduce the Hamiltonian, whose expression for all performance
indexes is
µ(s) T
H= λ2 w(x) s n + λT1 γs −n + λ3 µ(s) s n (B.37)
1 + Bµ(s)
where λ1 and λ2 (column vectors, with I + 1 elements) and λ3 are the Lagrange
multipliers defined by the following differential equations:
dλ1 µ(s)
=− s n [JT λ2 ] (B.38)
ds 1 + Bµ(s)
dλ2
= −λ1 s −n (B.39)
ds
dλ3
=0 (B.40)
ds
where J is the Jacobian matrix of the vector w with respect to the independent
variable x. The relevant BCs for the Lagrange multipliers are
s = 0: λ1 = 0 (B.41a)
−1 T
s = 1: λ2 = (A ) λ1 − JT1 λ1 +v (B.41b)
where J1 is the Jacobian matrix of p3 with respect to γ, and v is a column vector

of I + 1 elements which depends on the objective functional being maximized as
follows:
For conversion maximization
vi = 0 for i = N (B.42a)
KN
vN = . (B.42b)
QF0 ζF,N
0
+ QP0 ζP,N
0
For selectivity maximization
vi = 0 for i = N, D (B.43a)
QF0 ζF,D
0
+ QP0 ζP,D
0
− KDγD(1)
vN = − (B.43b)
KN γ N2 (1)
KD
vD = − . (B.43c)
KN γN (1)
For product purity maximization
v = dγ (B.44)
where d γ is the column vector of the partial derivatives of D with respect to γ.

Having formulated the problem, the necessary condition for optimality,
∂H
=0 (B.45)
∂µ
is derived in Appendix A.
Notation
A surface area of active element per unit catalyst pellet weight (m2 per gram of
catalyst pellet)
A matrix defined by equation (2.102)
Ai component involved in a reaction
a dimensionless catalyst activity
ap pore surface area per unit volume
B bq̄
Bi∗ dimensionless parameter defined by equation (2.32)
Bih Biot number for heat transfer, hR/λe
Bim Biot number for mass transfer, kg R/De
b constant defined by equation (2.122)
b1 , b2 constants defined by equation (3.27)
C concentration
c concentration vector
c capacitance
cp fluid heat capacity
D catalyst dispersion
De effective diffusivity
Di dimensionless effective diffusivity
D0 catalyst dispersion as loading q → 0
Da Damköhler number
d number of zones in a reactor
E activation energy
e reactant consumption rate defined by equation (3.11)
e elementary charge
F Faraday constant or flux
f, f j dimensionless reaction rate
f¯ dimensionless production rate of desired product, integrated over the
catalyst pellet
G∗ function defined by equation (2.80)
G performance index for deactivating catalyst defined by equation (4.12)
195
196 Notation
g dimensionless concentration in the external fluid phase

gv volume-averaged catalyst density
H Hamiltonian functional
H average reactant conversion over catalyst lifetime defined by equation (4.5)
h heat transfer coefficient or function defined by equations (2.137), (2.160), (2.187)
I ionic strength
I0 , I1 Bessel functions
J Jacobian matrix defined by equation (2.144)
Jc capillary flow rate
Jd diffusive transport rate
Jv evaporation rate
K equilibrium constant
K performance index for deactivating catalyst defined by equation (4.13)
Kg permeability of gel layer
Km Michaelis–Menten constant
Ksol equilibrium adsorption constant due to solution effects
Kst equilibrium adsorption constant due to steric hinderance
k local reaction rate constant
k̄ volume-averaged reaction rate constant
k+ adsorption rate constant
k− desorption rate constant
kB Boltzmann constant
kg mass transfer coefficient
ksol desorption rate constant due to solution effects
L reactor length
d[ ]
L[ ] differential operator: ( s1n ){ dsd (s n ds )}
Lg gel layer thickness
M maximum value of dimensionless reaction rate defined by equation (2.16)
m reaction order or constant defined by equation (7.3)
NA Avogadro’s number
Np number of pellets
Ns surface site density
NX surface area of oxide per unit solution volume
n integer, characteristic of pellet geometry: = 0 for infinite slab, = 1 for infinite
cylinder, = 2 for sphere; or surface concentration
P pressure
p specific surface area of active element (m2 per gram of active element)
Q molar flow rate
q local weight fraction of catalyst (grams of active element per gram of
catalyst pellet)
q̄ volume-averaged weight fraction of catalyst
R characteristic dimension: half thickness (n = 0), radius (n = 1, 2)
Rg universal gas constant
Rm reactant molecule radius
Rp pore radius
Notation 197
r, r j , r reaction rate
r̄ volume-averaged reaction rate defined by equation (3.3)
rf position of imbibition front
r0 reaction rate in a catalyst pellet where the active surface area per unit
catalyst weight is equal to pq̄
S selectivity
St Stanton number
s dimensionless radial coordinate
s̄ dimensionless location of catalyst in pellet
T temperature
t time
U1 , U2 objective functions defined by equations (2.108), (2.109)
u dimensionless solid-phase concentration
ū dimensionless concentration at the catalyst location
u dimensionless solid-phase concentration vector
V volume
Vp pellet volume
v velocity or bond valence
W catalyst weight or mass flux
w dimensionless feed distribution along a reactor
w vector defined by equation (2.101)
w Mx vector defined by equation (2.143)
X reactant conversion
XT total conversion for membrane reactor defined by equation (5.1)
x radial coordinate
x vector defined by equation (2.101)
xf vector defined by equation (2.101)
Y yield
y dimensionless axial coordinate
ŷ reactor portion with subsurface optimal location or reactor portion
with distributed feed
Z ion valence
z axial coordinate
[ ] concentration
Greek Letters
α upper bound of the dimensionless active-element concentration
defined by equation (2.128)
α1 weighting coefficient proportional to value of products
α2 weighting coefficient proportional to cost of raw materials
α3 weighting coefficient proportional to cost of catalyst
αg constant [see equation (5.14)]
β dimensionless heat of reaction or fraction of sulfonic groups
β̄ external Prater number
198 Notation
" dimensionless molar flow rate

γ dimensionless activation energy or perturbation [see equation (2.84)]
γ vector defined by equation (2.135)
dimensionless half thickness of step distribution
H heat of reaction
Ha heat of adsorption
Pg pressure drop in gel layer
T temperature rise
δ deposition thickness
δM membrane thickness
δ(s − s̄) Dirac-delta function at s̄
δa perturbation of catalyst distribution
δu perturbation of dimensionless concentration
δη perturbation of effectiveness factor
ε dimensionless heat of adsorption or void fraction
ε0 permittivity of vacuum
εr relative permittivity of solvent
ζ mole fraction
η effectiveness factor or viscosity
dimensionless time: for membrane reactor defined by equation (5.3)
θ dimensionless temperature
θ¯ dimensionless temperature at the catalyst location
κ ratio of Thiele moduli, φ22 /φ12
dimensionless parameter defined by equation (2.26)
e dimensionless parameter defined by equation (2.31)
e dimensionless parameter defined by equation (2.56)
λ Lagrange multiplier
λe effective thermal conductivity
µ dimensionless active element concentration, or Bim /Bih
νij stoichiometric coefficient of the ith component in the jth reaction
ξ dimensionless coordinate in a membrane
ξ̄ dimensionless catalyst location in a membrane
π empirical coefficient defined by equation (4.4)
ρ density
ρd deposited layer density
ρs pellet density
σ dimensionless adsorption equilibrium constant, or electric charge density
σc critical value of dimensionless adsorption equilibrium constant defined
by equation (2.47)
τ dimensionless temperature in the external fluid phase
υ volumetric flow rate
switching function
φ Thiele modulus
Notation 199
φ0 “clean” Thiele modulus defined by equation (2.24)

' electrostatic potential
√
ψ modified Thiele modulus, φ/ B
ψn function of catalyst location defined by equation (2.121)
function defined by equation (2.112)
ω function defined by equation (2.173)
Subscripts
0 0 plane; initial
b bulk liquid
c coolant
d diffuse layer
e equilibrium
ext external
F feed section of a membrane reactor
f fluid phase
IHP inner Helmholz plane
i component i
ip for product inhibition
is for substrate inhibition
j reaction j
M membrane
m, max maximum
OHP outer Helmholz plane
oper operating
opt optimal
P permeate
p poison or pore
r reactor
s saturation or surface
un uniform
Superscripts
0 reactor inlet
* value corresponding to the optimal catalyst distribution
References
Abello, M. C., Velasco, A. P., Gorriz, O. F., and Rivarola, J. B. 1995. Temperature-
programmed desorption study of the acidic properties of γ -alumina, Appl. Catal. A.
129: 93–100.
Aben, P. C. 1968. Palladium areas in supported catalysts. Determination of palladium sur-
face areas in supported catalysts by means of hydrogen chemisorption, J. Catal. 10:
224–229.
Adcock, D. S., and McDowall, I. C. 1957. The mechanism of filter pressing and slip casting,
J. Amer. Ceram. Soc. 40: 355–362.
Agashe, K. B., and Regalbuto, J. R. 1997. A revised physical theory for adsorption of metal
complexes at oxide surfaces, J. Coll. Interf. Sci. 185: 174–189.
Ahn, J. H., Ihm, S. K., and Park, K. S. 1988. The effects of the local concentration and
distribution of sulfonic acid groups on 1-butene isomerization catalyzed by macroporous
ion-exchange resin catalyst, J. Catal. 113: 434–443.
Akratopulu, K. C., Vordonis, L., and Lycourghiotis, A. 1986. Effect of temperature on
the point of zero charge and surface dissociation constants in aqueous suspensions of
γ -Al2 O3 , J. Chem. Soc., Faraday, Trans. I 82: 3697–3708.
Akratopulu, K. C., Vordonis, L., and Lycourghiotis, A. 1988. Development of carriers with
controlled concentration of charged surface groups in aqueous solutions. III. Regulation
of the point of zero charge, surface dissociation constants and concentration of charged
surface groups of SiO2 by variation of the solution temperature or by modification with
sodium ions, J. Catal. 109: 41–50.
Alexiou, M. S., and Sermon, P. A. 1993. Aspects of the preparation of heterogeneous
catalysts by impregnation, React. Kinet. Catal. Lett. 51: 1–7.
Anderson, J. R. 1975. Structure of Metallic Catalysts, London: Academic Press.
Arakawa, H., Takeuchi, K., Matsuzaki, T., and Sugi, Y. 1984. Effect of metal dispersion
on the activity and selectivity of Rh/SiO2 catalyst for high pressure CO hydrogenation,
Chem. Lett. 1607–1610.
Arbabi, S., and Sahimi, M. 1991a. Computer simulations of catalyst deactivation – I. Model
formulation and validation, Chem. Eng. Sci. 46: 1739–1747.
Arbabi, S., and Sahimi, M. 1991b. Computer simulations of catalyst deactivation – II. The
effect of morphological transport and kinetic parameters on the performance of the
catalyst, Chem. Eng. Sci. 46: 1749–1755.
Ardiles, D. R. 1986. Activity and selectivity of nonuniform bifunctional catalysts. Analysis
of the fixed-bed reactor performance, Collect. Czech. Chem. Commun. 51: 2509–2520.
Ardiles, D. R., Scelza, O. A., and Castro, A. A. 1985. Activity and selectivity of non-uniform
bifunctional catalysts, Collect. Czech. Chem. Commun. 50: 726–737.
201
202 References
Ardiles, D. R., De Miguel, S. R., Castro, A. A., and Scelza, O. A. 1986. Pt–Re catalysts:
Study of the impregnation step, Appl. Catal. 24: 175–186.
Aris, R. 1975. The Mathematical Theory of Diffusion and Reaction in Permeable Catalysts,
Oxford: Clarendon Press.
Asua, J. M., and Delmon, B. 1987. Theoretical study of the influence of nonuniform active-
phase distribution on activity and selectivity of hydrodesulfurization catalysts, Ind. Eng.
Chem. Res. 26: 32–39.
Au, S. S., Dranoff, J. S., and Butt, J. B. 1995. Nonuniform activity distribution in catalyst
particles: Benzene hydrogenation on supported nickel in a single pellet diffusion reactor,
Chem. Eng. Sci. 50: 3801–3812.
Bacaros, T., Bebelis, S., Pavlou, S., and Vayenas, C. G. 1987. Optimal catalyst distribution
in pellets with shell progressive poisoning: The case of linear kinetics. In Catalyst Deac-
tivation 1987, eds. Delmon, B., and Froment, G. F., pp. 459–468. Amsterdam: Elsevier.
Baiker, A., and Holstein, W. L. 1983. Impregnation of alumina with copper chloride –
modeling of impregnation kinetics and internal copper profiles, J. Catal. 84: 178–188.
Baiker, A., Monti, D., and Wokaun, A. 1986. Impregnation of alumina with copper chloride:
Evidence for differently immobilized copper species, Appl. Catal. 23: 425–436.
Baratti, R., Cao, G., Morbidelli, M., and Varma, A. 1990. Optimal activity distribution in
nonuniformly impregnated catalyst particles: Numerical analysis, Chem. Eng. Sci. 45:
1643–1646.
Baratti, R., Wu, H., Morbidelli, M., and Varma, A. 1993. Optimal catalyst activity profiles
in pellets – X. The role of catalyst loading, Chem. Eng. Sci. 48: 1869–1881.
Baratti, R., Gavriilidis, A., Morbidelli, M., and Varma, A. 1994. Optimization of a non-
isothermal nonadiabatic fixed-bed reactor using Dirac-type silver catalysts for ethylene
epoxidation, Chem. Eng. Sci. 49: 1925–1936.
Baratti, R., Feckova, V., Morbidelli, M., and Varma, A. 1997. Optimal catalyst activity
profiles in pellets. 11. The case of multiple-step distributions, Ind. Eng. Chem. Res. 36:
3416–3420.
Barbier, J., and Marecot, P. 1981. Comparative study of structure sensitive reactions on
Pt/Al2 O3 and Ir/Al2 O3 catalysts, Nouveau J. Chim. 5: 393–396.
Barrow, N. J., and Bowden, J. W. 1987. A comparison of models for describing the adsorption
of anions on a variable charge mineral surface, J. Coll. Interf. Sci. 119: 236–250.
Bartholomew, C. H. 1993. Sintering kinetics of supported metals: New perspectives from a
unifying GPLE treatment, Appl. Catal. A 107: 1–57.
Bartholomew, C. H., and Farrauto, R. J. 1976. Chemistry of nickel–alumina catalysts,
J. Catal. 45: 41–53.
Bartholomew, C. H., Pannell, R. B., and Butler, J. L. 1980. Support and crystallite size
effects in CO hydrogenation on nickel, J. Catal. 65: 335–347.
Barto, M., Markos, J., and Brunovska, A. 1991. Oscillatory behaviour of a catalyst pellet
with narrow region of activity, Chem. Eng. Sci. 46: 2875–2880.
Basset, J. M., Dalmai-Imelik, G., Primet, M., and Mutin, R. 1975. A study of ben-
zene hydrogenation and identification of the adsorbed species with Pt/Al2 O3 catalysts,
J. Catal. 37: 22–36.
Beck, D. D., Sommers, J. W., and DiMaggio C. L. 1997. Axial characterization of catalytic
activity in close-coupled lightoff and underfloor catalytic converters, Appl. Catal. B 11:
257–272.
Becker, E. R., and Nuttall, T. A. 1979. Controlled catalyst distribution on supports by co-
impregnation. In Preparation of Catalysts II, eds. Delmon, B., Grange, P., Jacobs, P., and
Poncelet, G., pp. 159–169. Amsterdam: Elsevier.
Becker, E. R., and Wei, J. 1977a. Nonuniform distribution of catalysts on supports. I.
Bimolecular Langmuir reactions, J. Catal. 46: 365–371.
Becker, E. R., and Wei, J. 1977b. Nonuniform distribution of catalysts on supports. II. First
order reactions with poisoning, J. Catal. 46: 372–381.
References 203
Beeckman, J. W., and Hegedus, L. L. 1991. Design of monolith catalysts for power plant
NOx emission control, Ind. Eng. Chem. Res. 30: 969–978.
Benesi, H. A., Curtis, R. M., and Studer, H. P. 1968. Preparation of highly dispersed catalytic
metals. Platinum supported on silica gel, J. Catal. 10: 328–335.
Benjamin, M. M., and Leckie, J. O. 1981. Multiple-site adsorption of Cd, Cu, Zn and Pb on
amorphous iron oxyhydroxide, J. Coll. Interf. Sci. 79: 209–221.
Berenblyum, A. S., Mund, S. L., Karelskii, V. V., Goranskaya, T. P., Zolotukhin, V. E., and
Lakhman, L. I. 1986. Catalysts for the selective hydrogenation of unsaturated compounds
consisting of alumina particles with a regular palladium distribution, Kinet. and Catal.
27: 184–187.
Berger, D., and Pei, D. C. T. 1973. Drying of hygroscopic capillary porous solids – a theo-
retical approach, Int. J. Heat Mass Transfer 16: 293–302.
Blachou, V. M., and Philippopoulos, C. J. 1993. Adsorption of hexachloroplatinic acid on γ -
alumina coatings for preparation of monolithic structure catalysts, Chem. Eng. Commun.
119: 41–53.
Blanco, M. N., Caceres, C. V., Fierro, J. L. G., and Thomas, H. J. 1987. Influence of the
operative conditions on the preparation of pelleted Mo/Al2 O3 catalysts, Appl. Catal. 33:
231–244.
Bolan, N. S., and Barrow, N. J. 1984. Modelling the effect of adsorption of phosphate
and other anions on the surface charge of variable charge oxides, J. Soil Sci. 35: 273–
281.
Bonivardi, A. L., and Baltanas, M. A. 1990. Preparation of Pd/SiO2 catalysts for methanol
synthesis, J. Catal. 125: 243–259.
Borchert, A., and Buchholz, K. 1979. Inhomogeneous distribution of fixed enzymes inside
carriers, Biotechnol. Lett. 1: 15–20.
Borchert, A., and Buchholz, K. 1984. Improved biocatalyst effectiveness by controlled
immobilization of enzymes, Biotechnol. and Bioeng. 26: 727–736.
Bothe, N. 1982. Struktur, Stabilität und katalytische Aktivität sulfonsaurer Styrol-
Divinylbenzol-Harze, PhD Thesis, Technische Universität zu Braunschweig.
Bournonville, J. P., Franck, J. P., and Martino, G. 1983. Influence of the various activation
steps on the dispersion and the catalytic properties of platinum supported on chlorinated
alumina. In Preparation of Catalysts III, eds. Poncelet, G., Grange P., and Jacobs, P. A.,
pp. 81–90. Amsterdam: Elsevier.
Bowden, J. W., Posner, A. M., and Quirk, J. P. 1977. Ionic adsorption on variable charge
mineral surfaces. Theoretical charge development and titration curves, Aust. J. Soil. Res.
15: 121–136.
Bowden, J. W., Nagarajah, S., Barrow, N. J., Posner, A. M., and Quirk, J. P. 1980. Describing
the adsorption of phosphate, citrate and selenite on a variable-charge mineral surface,
Aust. J. Soil. Res. 18: 49–60.
Brunelle, J. P. 1978. Preparation of catalysts by adsorption of metal complexes on mineral
oxides, Pure Appl. Chem. 50: 1211–1229.
Brunelle, J. P., Sugler, A., and Le Page, J.-F. 1976. Active centers of platinum silica catalysts
in hydrogenolysis and isomerization of n-pentane, J. Catal. 43: 273–291.
Brunovska, A. 1987. Dynamic behaviour of a catalyst pellet with nonuniform activity dis-
tribution, Chem. Eng. Sci. 42: 1969–1976.
Brunovska, A. 1988. Dynamic behaviour of catalyst pellet with step function activity dis-
tribution, Chem. Eng. Sci. 43: 2546–2548.
Brunovska, A., Morbidelli, M., and Brunovsky, P. 1990. Optimal catalyst pellet activity
distributions for deactivating systems, Chem. Eng. Sci. 45: 917–925.
Brunovska, A., Remiarova, B., and Pranda, P. 1994. Role of catalyst pellet activity distri-
bution in catalyst poisoning, Appl. Catal. A 108: 141–156.
Buchholz, K. 1979. Nonuniform enzyme distribution inside carriers, Biotechnol. Lett. 1:
451–456.
204 References
Buchholz, K., and Klein, J. 1987. Characterization of immobilized biocatalysts, Methods

Enzymol. 135: 3–30.
Burch, R., and Hayes, M. J. 1997. The preparation and characterisation of Fe-promoted
Al2 O3 -supported Rh catalysts for the selective production of ethanol from syngas, J.
Catal. 165: 249–261.
Butt, J. B., Downing, D. M., and Lee, J. W. 1977. Inter-intraphase temperature gradients in
fresh and deactivated catalyst particles, Ind. Eng. Chem. Fundam. 16: 270–272.
Caldwell, A. D., and Calderbank, P. H. 1969. Catalyst dilution – a means of temperature
control in packed tubular reactors, Brit. Chem. Eng. 14: 1199–1201.
Cannon, K. C., and Hacskaylo, J. J. 1992. Evaluation of palladium impregnation on the
performance of a Vycor glass catalytic membrane reactor, J. Membr. Sci. 65: 259–
268.
Carballo, L., Serrano, C., Wolf, E. E., and Carberry, J. J. 1978. Hydrogen chemisorption
studies on supported platinum using the flow technique, J. Catal. 52: 507–514.
Carberry, J. J. 1976. Chemical and Catalytic Reaction Engineering, New York: McGraw-Hill.
Carleysmith, S. W., Dunnill, P., and Lilly, M. D. 1980a. Kinetic behavior of immobilized
penicillin acylase, Biotechnol. and Bioeng. 22: 735–756.
Carleysmith, S. W., Eames, M. B. L., and Lilly, M. D. 1980b. Staining method for determi-
nation of the penetration of immobilised enzyme into a porous support, Biotechnol. and
Bioeng. 22: 957–967.
Castro, A. A., Scelza, O. A., Benvenuto, E. R., Baronetti, G. T., De Miguel, S. R., and Parera,
J. M. 1983. Competitive adsorption of H2 PtCl6 and HCl on Al2 O3 in the preparation of
naptha reforming catalysts. In Preparation of Catalysts III, eds. Poncelet, G., Grange, P.,
and Jacobs, pp. 47–56. Amsterdam: Elsevier.
Cervello, J., Hermana, E., Jimenez, J. F., and Melo, F. 1976. Effect of the impregnation
conditions on the internal distribution of the active species in catalysts. In Preparation
of Catalysts I, eds. Delmon, B., Jacobs, P. A., and Poncelet G., pp. 251–263. Amsterdam:
Elsevier.
Cervello, J., Melendo, J. F. J., and Hermana, E. 1977. Effect of variable specific rate contant
in nonuniform catalysts, Chem. Eng. Sci. 32: 155–159.
Charmas, R., Piasecki, W., and Rudzinski, W. 1995. Four layer complexation model for ion
adsorption at electrolyte/oxide interface: Theoretical foundations, Langmuir 11: 3199–
3210.
Chary, K. V. R., Rama Rao, B., and Subrahmanyam, V. S. 1991. Characterization of sup-
ported vanadium oxide catalysts by a low-temperature oxygen chemisorption technique
III. The V2 O5 /ZrO2 system, Appl. Catal. 74: 1–13.
Che, M., and Bennett, C. O. 1989. The influence of particle size on the catalytic properties
of supported metals, Adv. Catal. 36: 55–172.
Che, M., and Bonneviot, L. 1988. The change of properties of transition metal ions and the
role of support as a function of catalyst preparation. In Successful Design of Catalysts,
ed. Inui, T., pp. 147–158. Amsterdam: Elsevier.
Chee, Y. C., and Ihm, S. K. 1986. The influence of acid site distribution on the catalytic
deactivation of sulfonated poly(styrene–divinylbenzene) membrane catalyst, J. Catal.
102: 180–189.
Chemburkar, R. M. 1987. Optimal catalyst activity profiles in pellets. Single pellet theory
and experiments, PhD Thesis, University of Notre Dame.
Chemburkar, R. M., Morbidelli, M., and Varma, A. 1987. Optimal catalyst activity profiles
in pellets – VII. The case of arbitrary reaction kinetics with finite external heat and mass
transport resistances, Chem. Eng. Sci. 42: 2621–2632.
Chen, H.-C., and Anderson, R. B. 1973. Study of impregnated chromia on alumina cata-
lysts with an electron probe microanalyzer, Ind. Eng. Chem. Prod. Res. Dev. 12: 122–
127.
Chen, H.-C., and Anderson, R. B. 1976. Concentration profiles in impregnated chromium
and coppper on alumina, J. Catal. 43: 200–206.
References 205
Chen, H.-C., Gillies, G. C., and Anderson, R. B. 1980. Impregnating chromium and copper
in alumina, J. Catal. 62: 367–373.
Cheng, W. C., and Pereira, C. J. 1987. Preparation of cobalt–molybdenum hydrotreating
catalysts using hydrogen peroxide stabilization, Appl. Catal. 33: 331–341.
Chiang, C.-L., and Fang, Z.-R. 1994. Simulation and optimal design for the resid-
ual oil hydrodemetallation in a cocurrent moving-bed reactor, Chem. Eng. Sci. 49:
1175–1183.
Chiang, C.-L., and Tiou, H.-H. 1992. Optimal design for the residual oil hydrodemetallation
in a fixed bed reactor, Chem. Eng. Commun. 117: 383–399.
Chibata, I., Tosa, T., and Shibatani, T. 1992. The industrial production of optically active
compounds by immobolized biocatalysts. In Chirality in Industry, eds. Collins, A. N.,
Sheldrake, G. N., and Crosby, J., pp. 352–370. Wiley.
Chu, P., Petersen, E. E., and Radke, C. J. 1989. Modeling wet impregnation of nickel on γ -
alumina, J. Catal. 117: 52–70.
Chung, B. H., and Chang, H. N. 1986. Effect of internal diffusion on the apparent stability
of nonuniformly distributed biocatalysts, Korean J. Chem. Eng. 3: 39–43.
Clause, O., Kermarec, M., Bonneviot, L., Villain, F., and Che, M. 1992. Nickel(II) ion-
support interactions as a function of preparation method of silica-supported nickel ma-
terials, J. Am. Chem. Soc. 114: 4709–4717.
Contescu, C., and Vass, M. I. 1987. The effect of pH on the adsorption of palladium(II)
complexes on alumina, Appl. Catal. 33: 259–271.
Contescu, C., Hu, J., and Schwarz, J. A. 1993a. 1-pK multisites description of charge devel-
opment at the aqueous alumina interface, J. Chem. Soc. Faraday Trans. 89: 4091–4099.
Contescu, C., Jagiello, J., and Schwarz, J. A. 1993b. Heterogeneity of proton binding sites
at the oxide/solution interface, Langmuir 9: 1754–1765.
Contescu, C., Macovei, D., Craiu, C., Teodorescu, C., and Schwarz, J. A. 1995a. Thermal
induced evolution of chlorine-containing precursors in impregnated Pd/Al2 O3 catalysts,
Langmuir 11: 2031–2040.
Contescu, C., Jagiello, J., and Schwarz, J. A. 1995b. Proton affinity distributions: A scientific
basis for the design and construction of supported metal catalysts. In Preparation of
Catalysts VI, eds. Poncelet, G., Martens, J., Delmon, B., Jacobs, P. A., and Grange, P., pp.
237–252. Amsterdam: Elsevier.
Corbett, W. E., and Luss, D. 1974. The influence of non-uniform catalytic activity on the
performance of a single spherical pellet, Chem. Eng. Sci. 29: 1473–1483.
Cot, L., Guizard, C., and Larbot, A. 1988. Novel ceramic material for liquid separation
process: Present and prospective applications in microfiltration and ultrafiltration, Ind.
Ceram. 8: 143–148.
Coulson, J. M., and Richardson, J. F. 1991. Drying. In Chemical Engineering, 4th ed., vol.
2, Chapter 16. Oxford: Pergamon.
Cukierman, A. L., Laborde, M. A., and Lemcoff, N. O. 1983. Optimum activity distribution
in a catalyst pellet for a complex reaction, Chem. Eng. Sci. 38: 1977–1982.
Cusumano, J. A. 1991. Creating the future of the chemical industry – catalysts by molecular
design. In Perspectives in Catalysis, eds. Thomas, J. M., and Zamaraev, K. I., pp. 1–33.
Oxford: Blackwell Scientific.
Dadyburjor, D. B. 1982. Distribution for maximum activity of a composite catalyst,
A.I.Ch.E.J. 28: 720–728.
Dadyburjor, D. B. 1985. Selectivity over unifunctional multicomponent catalysts with
nonuniform distribution of components, Ind. Eng. Chem. Fundam. 24: 16–27.
Dadyburjor, D. B., and White, C. W., III 1990. Effect of position-dependent deactivation
on the design of a composite cracking catalyst, Chem. Eng. Sci. 45: 2619–2624.
D’Aniello, M. J. 1981. Anion adsorption on alumina, J. Catal. 69: 9–17.
Davis, J. A., and Leckie, J. O. 1978. Surface ionization and complexation at the oxide/water
interface 2. Surface properties of amorphous iron oxyhydroxide and adsorption of metal
ions, J. Coll. Interf. Sci. 67: 90–107.
206 References
Davis, J. A., and Leckie, J. O. 1980. Surface ionization and complexation at the oxide/water
interface 3. Adsorption of anions, J. Coll. Interf. Sci. 74: 32–43.
Davis, J. A., James, R. O., and Leckie J. O. 1978. Surface ionization and complexation at
the oxide/water interface 1. Computation of electrical double layer properties in simple
electrolytes, J. Coll. Interf. Sci. 63: 480–499.
De Jong, K. P. 1991. Deposition precipitation onto pre-shaped carrier bodies. Possibilities
and limitations. In Preparation of Catalysts V, eds. Poncelet, G., Jacobs, P. A., Grange, P.,
and Delmon, B., pp. 19–36. Amsterdam: Elsevier.
DeLancey, G. B. 1973. An optimal catalyst activation policy for poisoning problems, Chem.
Eng. Sci. 28: 105–118.
Delannay, F. (ed.) 1984. Characterization of Heterogeneous Catalysts. New York: Marcel
Dekker.
Delmon, B., and Houalla, M. 1979. Tentative classification of the factors influencing the
reduction step in the activation of supported catalysts. In Preparation of Catalysts II, eds.
Delmon, B., Grange, P., Jacobs, P., and Poncelet, G., pp. 439–468. Amsterdam: Elsevier.
De Miguel, S. R., Scelza, O. A., Castro, A. A., Baronetti, G. T., Ardiles, D. R., and Parera,
J. M. 1984. Radial profiles in Pt/Al2 O3 , Re/Al2 O3 , and Pt–Re/Al2 O3 , Appl. Catal. 9:
309–315.
D’Espinose de la Caillerie, J. -B., Bobin, C., Rebours, B., and Clause, O. 1995. Alumina/water
interfacial phenomena during impregnation. In Preparation of Catalysts VI, eds. Poncelet,
G., Martens, J., Delmon, B., Jacobs, P., and Grange, P., pp. 169–184. Amsterdam: Elsevier.
Do, D. D. 1984. A method for solving diffusion and reaction problems with nonuniform
activity catalysts, Chem. Eng. Sci. 39: 1519–1522.
Do, D. D. 1985. Modeling of impregnation kinetics and internal activity profiles: Adsorption
of HCl, HReO4 and H2 PtCl6 onto γ -Al2 O3 . Chem. Eng. Sci. 40: 1871–1880.
Do, D. D., and Bailey, J. E. 1982. Approximate analytical solutions for porous catalysts with
nonuniform activity, Chem. Eng. Sci. 37: 545–551.
Dorling, T. A., Lynch, B. W. J., and Moss, R. L. 1971. The structure and activity of supported
metal catalysts. V. Variables in the preparation of platinum/silica catalysts, J. Catal. 20:
190–201.
Dorling, T. A., and Moss, R. L. 1967. The structure and activity of supported metal catalysts.
II. Crystallite size and CO chemisorption in platinum/silica catalysts, J. Catal. 7: 378–385.
Dougherty, R. C., and Verykios, X. E. 1987. Nonuniformly activated catalysts, Catal. Rev. –
Sci. Eng. 29: 101–150.
Downing, D. M., Lee, J. W., and Butt, J. B. 1979. Simulation models for intraparticle deac-
tivation: Scope and reliability, A.I.Ch.E.J. 25: 461–469.
Duncombe, P. R., and Weller, S. W. 1985. Thermal behavior of shell molybdena–alumina
catalysts, A.I.Ch.E.J. 31: 410–414.
Engels, S., Lausch, H., and Schwokowski, R. 1987. Untersuchungen zur Adsorption bi- und
polyfunktioneller organischer Saeuren an Aluminiumoxid, Chem. Techn. (Leipzig) 39:
387–391.
Ernst, W. R., and Daugherty, D. J. 1978. Method for the study of performance of a single
spherical particle with nonuniform catalytic activity, A.I.Ch.E.J. 24: 935–937.
Fenelonov, V. B. 1975. Analysis of the drying stages in the technology of deposited catalysts
I. Role of the pore structure and the drying conditions, Kinet. and Catal. 16: 628–635.
Fenelonov, V. B., Neimark, A. V., and Kheifets, L. I. 1979. Analysis of steps of impreg-
nation and drying in preparation of supported catalysts. In Preparation of Catalysts
II, eds. Delmon, B., Grange, P., Jacobs, P., and Poncelet, G., pp. 233–244. Amsterdam:
Elsevier.
Finlayson, B. A. 1980. Nonlinear Analysis in Chemical Engineering. New York: McGraw-
Hill.
Foger, K. 1984. Dispersed metal catalysts. In Catalysis Science and Technology, vol. 6,
eds. Anderson, J. R., and Boudart, M., pp. 227–305. Berlin: Springer-Verlag.
References 207
Frost, A. C., Sawyer, J. E., Summers, J. C., Shah, Y. T., and Dassori, C. G. 1990. Reac-
tor model for the fixed bed catalytic incineration of VOC’s, presented at the 83rd An-
nual Meeting and Exhibition, Air and Waste Management Association, Pittsburgh, PA,
June 24–29.
Fuentes, S., and Figueras, F. 1980. The influence of particle size on the catalytic properties
of alumina-supported rhodium catalysts, J. Catal. 61: 443–453.
Fujiyama, T., Otsuka, M., Tsuiki, H., and Ueno, A. 1987. Control of the impregnation
profile of Ni in an Al2 O3 sphere, J. Catal. 104: 323–330.
Furusaki, S. 1988. Engineering aspects of immobilised biocatalysts, J. Chem. Eng. Japan 21:
219–230.
Galiasso, R., de Ochoa, O. L., and Andreu, P. 1983. Influence of the drying rate on the
distribution of active metals and on the activity of heavy crude hydrotreating catalyst,
Appl. Catal. 5: 309–322.
Gates, G. C., Winouskas, J. S., and Heath, H. W. 1972. The dehydration of t-butyl alcohol
catalyzed by sulfonic acid resin, J. Catal. 24: 320–327.
Gavriilidis, A., and Varma, A. 1992. Optimal catalyst activity profiles in pellets: 9. Study of
ethylene epoxidation, A.I.Ch.E.J. 38: 291–296.
Gavriilidis, A., Varma, A., and Morbidelli, M. 1993a. Optimal distribution of catalyst in
pellets, Catal. Rev. – Sci. Eng. 35: 399–456.
Gavriilidis, A., Sinno, B., and Varma, A. 1993b. Influence of loading on metal surface area
for Ag/α-Al2 O3 catalysts, J. Catal. 139: 41–47.
Gazimzyanov, N. R., Mikhailov, V. I., and Zadko, I. I. 1992. Study of distribution of active
components in alumino–nickel–molybdenum catalysts obtained by a single impregna-
tion, Kinet. and Catal. 33: 737–743.
Gelosa, D., Pedeferri, M. P., and Morbidelli, M. 1998. Esterification reactions with nonuni-
formly sulfonated resins, First Annual Report INTREASEP Project, Joule Contract
JOE3-CT97-0082.
Gladden, L. F. 1995. Applications of nuclear magnetic resonance imaging in particle tech-
nology, Part. Part. Syst. Charact. 12: 59–67.
Gottifredi, J. C., Gonzo, E. E., and Quiroga, O. D. 1981. Isothermal effectiveness factor II.
Analytical expression for single reaction with arbitrary kinetics, geometry and activity
distribution, Chem. Eng. Sci. 36: 713–719.
Goula, M. A., Kordulis, C., Lycourghiotis, A., and Fierro, J. L. G. 1992a. Development of
molybdena catalysts supported on γ -alumina extrudates with four different Mo profiles:
Preparation, characterisation and catalytic properties, J. Catal. 137: 285–305.
Goula, M. A., Kordulis, C., and Lycourghiotis, A. 1992b. Influence of impregnation param-
eters on the axial Mo/γ -alumina profiles studied using a novel simple technique, J. Catal.
133: 486–497.
Guisan, J. M., Serrano, J., Melo, F. V., and Ballesteros, A. 1987. Mixed enzymic reaction–
internal diffusion kinetics of nonuniformly distributed immobilised enzymes. The system
agarose–micrococcal endonuclease, Appl. Biochem. Biotech. 14: 49–72.
Hachiya, K., Sasaki, M., Saruta, Y., Mikami, N., and Yasunaga, T. 1984. Static and kinetic
studies of adsorption–desorption of metal ions on a γ -Al2 O3 surface. 1. Static study of
adsorption–desorption, J. Phys. Chem. 88: 23–27.
Hamada, H. Funaki, R., Kuwahara, Y., Kintaichi, Y., Wakabayashi, K., and Ito, T. 1987.
Systematic preparation of supported Rh catalysts having desired metal particle size by
using silica supports with controlled pore structure, Appl. Catal. 30: 177–180.
Hanika, J., and Ehlova, V. 1989. Modelling of internal diffusion inside a catalyst particle with
non-uniform radial activity profile for parallel reactions, Collect. Czech. Chem. Commun.
54: 81–90.
Hanika, J., Machek, V., Nemec, V., Ruzicka, V., and Kunz, J. 1982. Simultaneous diffusion
and adsorption of chloroplatinic acid in charcoal pellet during preparation process of
supported platinum catalyst, J. Catal. 77: 248–256.
208 References
Hanika, J., Hao, L. H., and Ruzicka, V. 1983. Relation between the conditions of preparation
and the activity of supported platinum catalysts, Collect. Czech. Chem. Commun. 48:
3079–3085.
Hanika, J., Janousek, V., and Sporka, K. 1987a. Modelling of impregnation of γ -alumina
with cobalt and molybdenum salts. CoCl2 –(NH4 )2 MoO4 –γ -Al2 O3 aluminate type sys-
tem, Collect. Czech. Chem. Commun. 52: 663–671.
Hanika, J., Janousek, V., and Sporka, K. 1987b. Modelling of impregnation of γ -alumina
with cobalt and molybdenum salts. Co(NO3 )2 –(NH4 )2 MoO4 –γ -Al2 O3 chloride type sys-
tem, Collect. Czech. Chem. Commun. 52: 672–677.
Harold, M. P., and Luss, D. 1987. Impact of the catalytic activity profile on observed multi-
plicity features: CO oxidation on Pt/Al2 O3 , Ind. Eng. Chem. Res. 26: 1616–1621.
Harriott, P. 1969. Diffusion effects in the preparation of impregnated catalysts, J. Catal. 14:
43–48.
Haworth, A. 1990. A review of the modelling of sorption from aqueous solution, Adv. Coll.
Interf. Sci. 32: 43–78.
Hayes, K. F., and Leckie, J. O. 1986. Mechanism of lead ion adsorption at the goethite–water
interface, ACS Symp. Ser. 323: 114–141.
Heck, R. M., and Farrauto, R. J. 1995. Catalytic Air Pollution Control. Commercial Tech-
nology, New York: Van Nostrand Reinhold.
Hegedus, L. L. 1980. Catalyst pore structures by constrained nonlinear optimisation, Ind.
Eng. Chem. Prod. Res. Dev. 19: 533–537.
Hegedus, L. L., and Gumbleton, J. J. 1980. Catalysts, computers and cars: A growing sym-
biosis, Chemtech, October, 630–642.
Hegedus, L. L., and McCabe, R. 1984. Catalyst Poisoning. New York: Marcel Dekker.
Hegedus, L. L., and Pereira, C. J. 1990. Reaction engineering for catalyst design, Chem.
Eng. Sci. 45: 2027–2044.
Hegedus, L. L., and Summers, J. C. 1977. Improving the poison resistance of supported
catalysts, J. Catal. 48: 345–353.
Hegedus, L. L., Summers, J. C., Schlatter, J. C., and Baron, K. 1979a. Poison resistant
catalysts for the simultaneous control of hydrocarbon, carbon monoxide and nitrogen
oxide emissions, J. Catal. 56: 321–335.
Hegedus, L. L., Chou, T. S., Summers, J. C., and Potter, N. M. 1979b. Multicomponent
chromatographic processes during the impregnation of alumina pellets with noble metals.
In Preparation of Catalysts II, eds. Delmon, B., Grange, P., Jacobs, P., and Poncelet, G.,
pp. 171–184. Amsterdam: Elsevier.
Heise, M. S., and Schwarz, J. A. 1985. Preparation of metal distributions within catalyst
supports I. Effect of pH on catalytic metal profiles, J. Coll. Interf. Sci. 107: 237–243.
supports II. Effect of ionic strength on catalytic metal profiles, J. Coll. Interf. Sci. 113:
55–61.
supports. In Preparation of Catalysts IV, eds. Delmon, B., Grange, P., Jacobs, P. A., and
Poncelet, G., pp. 1–13. Amsterdam: Elsevier.
supports III. Single component modeling of pH, ionic strength and concentration effects,
J. Coll. Interf. Sci. 123: 51–58.
Hepburn, J. S., Stenger, H. G., and Lyman, C. E. 1989a. Distributions of HF co-impregnated
rhodium, platinum and palladium in alumina honeycomb supports, Appl. Catal. 55: 271–
285.
Hepburn, J. S., Stenger, H. G., and Lyman, C. E. 1989b. Effects of drying on the preparation
of HF co-impregnated rhodium/Al2 O3 catalysts, Appl. Catal. 55: 287–299.
Hepburn, J. S., Stenger, H. G., and Lyman, C. E. 1989c. Co-impregnation of rhodium into
alumina honeycombs with acids and salts, Appl. Catal. 56: 107–118.
References 209
Hepburn, J. S., Stenger, H. G., and Lyman, C. E. 1991a. Co-impregnated Rh/Al2 O3 . I.

Preparation, J. Catal. 128: 34–47.
Hepburn, J. S., Stenger, H. G., and Lyman, C. E. 1991b. Co-impregnated Rh/Al2 O3 . II.
Nitric oxide reduction and SO2 poisoning, J. Catal. 128: 48–62.
Hepburn, J. S., Stenger, H. G., and Lyman, C. E. 1991c. Co-impregnation of rhodium chlo-
ride with hydrofluoric acid into dry and pre-wet alumina, Appl. Catal. 71: 205–218.
Hiemstra, T., van Riemsdijk, W. H., and Bolt, G. H. 1989a. Multisite proton adsorption mod-
eling at the solid/solution interface of hydroxides: A new approach. I. Model description
and evaluation of intrinsic reaction constants, J. Coll. Interf. Sci. 133: 91–104.
Hiemstra, T., de Wit, J. C. M., and van Riemsdijk, W. H. 1989b. Multisite proton adsorption
modeling at the solid/solution interface of hydroxides: A new approach. II. Application
to various important hydroxides, J. Coll. Interf. Sci. 133: 105–117.
Hoekstra, J. 1968. Catalyst for treatment of combustible waste products, U.S. Patent
3,388,077.
Hohl, H., and Stumm, W. 1976. Interaction of Pb2+ with hydrous γ -Al2 O3 , J. Coll. Interf.
Sci. 55: 281–288.
Hollewand, M. P., and Gladden, L. F. 1992. Visualization of phases in catalyst pellets and
pellet mass transfer processes using magnetic resonance imaging, Trans. IChemE 70A:
183–185.
Hollewand, M. P., and Gladden, L. F. 1994. Probing the structure of porous pellets: An
NMR study of drying, Magn. Reson. Imaging, 12, 291–294.
Horvath, C., and Engasser, J.-M. 1973. Pellicular heterogeneous catalysts. A theoretical
study of the advantages of shell structured immobilized enzyme particles, Ind. Eng.
Chem. Fundam. 12: 229–235.
Hossain, M. M., and Do, D. D. 1987. Effects of nonuniform immobilised enzyme distribution
in porous solid supports on the performance of a continuous reactor, Chem. Eng. J. 34:
B35–B47.
Hsieh, H. P. 1991. Inorganic membrane reactors, Catal. Rev. Sci. Eng. 33: 1–70.
Hu, Z., Wan, C. Z., Lui, Y. K., Dettling, J., and Steger, J. J. 1996. Design of a novel Pd
three-way catalyst: Integration of catalytic functions in three dimensions, Catal. Today
30: 83–89.
Huang, C. P. 1975. Adsorption of phosphate at the hydrous γ -Al2 O3 –electrolyte interface,
J. Coll. Interf. Sci. 53: 178–186.
Huang, Y.-J. R., Barrett, B. T., and Schwarz, J. A. 1986. The effect of solution variables on
metal weight loading during catalyst preparation, Appl. Catal. 24: 241–248.
Huang, Y.-J., and Schwarz, J. A. 1987. The effect of catalyst preparation on catalytic activity:
III. The catalytic activity of Ni/Al2 O3 catalysts prepared by incipient wetness, Appl. Catal.
32: 45–57.
Iannibello, A., and Mitchell, P. C. H. 1979. Preparative chemistry of cobalt–molybnenum/
alumina catalysts. In Preparation of Catalysts II, eds. Delmon, B., Grange, P., Jacobs, P.,
and Poncelet, G., pp. 469–478. Amsterdam: Elsevier.
Iannibello, A., Marengo, S., Trifiro, F., and Villa, P. L. 1979. A study of the chemisorp-
tion of chromium(VI), molybdenum(VI) and tungsten(VI) onto γ -alumina. In Prepara-
tion of Catalysts II, eds. Delmon, B., Grange, P., Jacobs, P., and Poncelet, G., pp. 65–76.
Amsterdam: Elsevier.
Iglesia, E. 1997. Design, synthesis, and use of cobalt-based Fischer-Tropsch synthesis cata-
lysts, Appl. Catal. A. 161: 59–78.
Iglesia, E., Reyes, S. C., Madon, R. J., and Soled, S. L. 1993. Selectivity control and catalyst
design in the Fischer–Tropsch synthesis: Sites, pellets and reactors, Adv. Catal. 39: 221–
302.
Iglesia, E., Soled, S. L., Baumgartner, J. E., and Reyes, S. C. 1995. Synthesis and catalytic
properties of eggshell cobalt catalysts for the Fischer–Tropsch synthesis, J. Catal. 153:
108–122.
210 References
Jianguo, W., Jiayu, Z., and Li, P. 1983. The role of competitive adsorbate in the impregnation
of platinum in pelleted alumina support. In Preparation of Catalysts III, eds. Poncelet,
G., Grange P., and Jacobs, P. A., pp. 57–67. Amsterdam: Elsevier.
Jiratova, K. 1981. Isoelectric point of modified alumina, Appl. Catal. 1: 165–167.
Johnson, D. L., and Verykios, X. E. 1983. Selectivity enhancement in ethylene oxidation
employing partially impregnated catalysts, J. Catal. 79: 156–163.
Johnson, D. L., and Verykios, X. E. 1984. Effects of radially nonuniform distributions of
catalytic activity on performance of spherical catalyst pellets, A.I.Ch.E.J. 30: 44–50.
Juang, H.-D., and Weng, H.-S. 1983. Performance of catalysts with nonuniform activity
profiles 2. Theoretical analysis for nonisothermal reactions. Ind. Eng. Chem. Fundam.
22: 224–230.
Juang, H.-D., and Weng, H.-S. 1984. Performance of biocatalysts with nonuniformly dis-
tributed immobilized enzyme, Biotechnol. and Bioeng. 26, 623–626.
Juang, H.-D., Weng, H.-S. and Wang, C.-C. 1981. The performance of catalysts with nonuni-
form activity profile I. Theoretical analysis for isothermal reactions, Stud. Surf. Sci. Catal.
7: 866–876.
Karakonstantis, L., Kordulis, Ch., and Lycourghiotis, A. 1992. Mechanism of adsorption
of tungstates on the interface of γ -alumina/electrolyte solutions, Langmuir 8: 1318–
1324.
Karpe, P., and Ruckenstein, E. 1989. Role of electrokinetic phenomena in the preparation
of supported metal catalysts, Coll. and Polym. Sci. 267: 145–150.
Kasaoka, S., and Sakata, Y. 1968. Effectiveness factors for non-uniform catalyst pellets,
J. Chem. Eng. Japan 1: 138–142.
Kehoe, P. 1974. Letter to the Editors, Chem. Eng. Sci. 29: 315.
Kehoe, J. P. G., and Butt, J. B. 1972. Interactions of inter- and intraphase gradients in a
diffusion limited catalytic reaction, A.I.Ch.E.J. 18: 347–355.
Keil, F. J., and Rieckmann, C. 1994. Optimisation of three-dimensional catalyst pore struc-
tures, Chem. Eng. Sci. 49: 4811–4822.
Keller, J. B., Falkovitz, M. S., and Frisch, H. 1984. Optimal catalyst distribution in a mem-
brane, Chem. Eng. Sci. 39: 601–604.
Kempling, J. C., and Anderson, R. B. 1970. Hydrogenolysis of n-butane on supported
ruthenium, Ind. Eng. Chem. Proc. Des. Dev. 9: 116–120.
Kheifets, L. I., Neimark, A. V., and Fenelonov, V. B. 1979. Analysis of the influence of
permeation and drying stages in supported catalyst preparation on the distribution of
components I. Precipitation of one component from solution, Kinet. and Catal. 20: 626–
632.
Kim, D. S., Kurusu, Y., Wachs, I. E., Hardcastle, F. D., and Segawa, K. 1989. Physicochemical
properties of MoO3 –TiO2 prepared by an equilibrium adsorption method, J. Catal. 120:
325–336.
Kimura, H. 1993. Selective oxidation of glycerol on a platinum–bismuth catalyst by using
a fixed bed reactor, Appl. Catal. A 105: 147–158.
King, D. L. 1978. A Fischer–Tropsch study of supported ruthenium catalysts, J. Catal. 51:
386–397.
Klein, J., Widdecke, H., and Bothe, N. (1984). Influence of functional group distribution
on the thermal stability and catalytic activity of sulfonated styrene-DVB-copolymers,
Macromol. Chem. and Phys., Suppl. 6: 211–226.
Klugherz, P. D., and Harriott, P. 1971. Kinetics of ethylene oxidation on a supported silver
catalyst, A.I.Ch.E.J. 17: 856–866.
Ko, S. H., and Chou, T. C. 1994. Hydrogenation of (–)-α-pinene over nickel–
phosphorus/aluminum oxide catalysts prepared by electroless deposition, Can. J. Chem.
Eng. 72: 862–873.
Komiyama, M. 1985. Design and preparation of impregnated catalysts, Catal. Rev. – Sci.
Eng. 27: 341–372.
References 211
Komiyama, M., and Muraki, H. 1990. Design of intrapellet activity profiles in supported
catalysts. In Handbook of Heat and Mass Transfer, vol. 4, ed. Cheremisinoff, N. P.,
pp. 447–500. Houston: Gulf Publishing Company.
Komiyama, M., Ohashi, K., Morioka, Y., and Kobayashi, J. 1997. Effects of the intrapellet
activity profiles on the selectivities in consecutive reactions: 1,3-butadiene hydrogenation
over Pt/Al2 O3 , Bull. Chem. Soc. Japan 70: 1009–1013.
Komiyama, M., Merrill, R. P., and Harnsberger, H. F. 1980. Concentration profiles in im-
pregnation of porous catalysts: Nickel on alumina, J. Catal. 63: 35–52.
Kotter, M., and Riekert, L. 1979. The influence of impregnation, drying and activation on
the activity and distribution of CuO on α-alumina. In Preparation of Catalysts II, eds.
Delmon, B., Grange, P., Jacobs, P., and Poncelet, G., pp. 51–63. Amsterdam: Elsevier.
Kotter, M., and Riekert, L. 1983a. Impregnation-type catalysts with nonuniform distribu-
tion of the active component. Part I: Influence of the accessibility of the active component
on activity and selectivity, Chem. Eng. Fundam. 2: 22–30.
Kotter, M., and Riekert, L. 1983b. Impregnation-type catalysts with nonuniform distri-
bution of the active component. Part II: Preparation and properties of catalysts with
different distribution of the active component on inert carriers, Chem. Eng. Fundam. 2:
31–38.
Kresge, C. T., Chester, A. W., and Oleck, S. M. 1992. Control of metal radial profiles in
alumina supports by carbon dioxide, Appl. Catal. A 81: 215–226.
Kulkarni, S. S., Mauze. G. R., and Schwarz, J. A. 1981. Concentration profiles and the design
of metal-supported catalysts, J. Catal. 69: 445–453.
Kummer, J. T. 1980. Catalysts for automobile emission control, Prog. Energy Combust. Sci.
6: 177–199.
Kummert, R., and Stumm, W. 1980. The surface complexation of organic acids on hydrous
γ -Al2 O3 , J. Colloid Interface Sci. 75: 373–385.
Kunimori, K., Nakajima, I., and Uchijima, T. 1982. Catalytic performance of egg white type
Pt/Al2 O3 catalyst in the oxidation of C3 H8 and CO, Chem. Lett. 1165–1168.
Kunimori, K., Kawasaki, E., Nakajima, I., and Uchijima, T. 1986. Catalytic performance of
egg-white type Pt/Al2 O3 catalyst: Multiple steady states in the oxidation of CO, Appl.
Catal. 22: 115–122.
Lee, C. K., and Varma, A. 1987. Steady state multiplicity behavior of an adiabatic plug-flow
reactor with nonuniformly active catalyst, Chem. Eng. Commun. 58: 287–309.
Lee, C. K., and Varma, A. 1988. An isothermal fixed-bed reactor with nonuniformly active
catalysts: Experiments and theory, Chem. Eng. Sci. 43: 1995–2000.
Lee, C. K., Morbidelli, M., and Varma, A. 1987a. Optimal catalyst activity profiles in pellets
6. Optimization of the isothermal fixed-bed reactor with multiple zones, Ind. Eng. Chem.
Res. 26: 167–170.
Lee, C. K., Morbidelli, M., and Varma, A. 1987b. Steady state multiplicity behavior of an
isothermal axial dispersion fixed-bed reactor with nonuniformly active catalyst, Chem.
Eng. Sci. 42: 1595–1608.
Lee, H. H. 1981. Generalized effectiveness factor for pellets with nonuniform activity dis-
tribution: Asymptotic region of strong diffusion effect, Chem. Eng. Sci. 36: 1921–1925.
Lee, H. H. 1984. Catalyst preparation by impregnation and activity distribution, Chem.
Eng. Sci. 39: 859–864.
Lee, H. H. 1985. Catalyst preparation by impregnation and activity distributions – II. Coim-
pregnation for nonuniform distribution based on competitive adsorption, Chem. Eng.
Sci. 40: 1295–1300.
Lee, H. H., and Ruckenstein, E. 1983. Catalyst sintering and reactor design, Cat. Rev. – Sci.
Eng. 25: 475–550.
Lee, J. W., Butt, J. B., and Downing, D. M. 1978. Kinetic, transport and deactivation rate in-
teractions on steady state and transient responses in heterogeneous catalysis, A.I.Ch.E.J.
24: 212–222.
212 References
Lee, S. H., and Ruckenstein, E. 1986. Optimum design of zeolite/silica–alumina catalysts,

Chem. Eng. Commun. 46: 43–64.
Lee, S.-Y., and Aris, R. 1985. The distribution of active ingredients in supported catalysts
prepared by impregnation, Catal. Rev. – Sci. Eng. 27: 207–340.
Le Page, J.-F. 1987. Applied Heterogeneous Catalysis: Design, Manufacture, Use of Solid
Catalysts, Paris: Editions Technip.
Letkova, Z., Brunovska, A., and Markos, J. 1994. Study of activity distribution for consec-
utive reactions, Collect. Czech. Chem. Commun. 59: 1788–1799.
Li, B., Liang, J., Han, B., and Liu, Y. 1985. The distribution control of active components
on support, J. Catal. (Chinese) 6: 95–99.
Li, W. D., Li, Y. W., Qin, Z. F., and Chen, S. Y. 1994. Theoretical prediction and experimental
validation of the egg-shell distribution of Ni for supported Ni/Al2 O3 catalysts, Chem. Eng.
Sci. 49: 4889–4895.
Lilly, M. D. 1996. Biotransformations using immobilised biocatalysts – past, present
and future. In Advances in Molecular and Cell Biology, vol. 15A, ed. Danielsson, B.,
pp. 139–147. London: JAI Press.
Limbach, K. W., and Wei, J. 1988. Effect of nonuniform activity on hydrodemetallation
catalyst, A.I.Ch.E.J. 34: 305–313.
Lin, T.-B., and Chou, T.-C. 1994. Selective hydrogenation of isoprene on eggshell and
uniform palladium profile catalysts, Appl. Catal. A 108: 7–19.
Lin, T.-B., and Chou, T.-C. 1995. Pd migration. 1. A possible reason for the deactivation of
pyrolysis gasoline partial hydrogenation catalysts, Ind. Eng. Chem. Res. 34: 128–134.
Lundquist, E. G. 1995. Catalysed esterification process, U.S. Patent 5,426,199.
Lycourghiotis, A. 1995. Preparation of supported catalysts by equilibrium deposition–
filtration. In Preparation of Catalysts VI, eds. Poncelet, G., Martens, J., Delmon, B.,
Jacobs, P. A., and Grange, P., pp. 95–129. Amsterdam: Elsevier.
Lyman, C. E., Hepburn, J. S., and Stenger, H. G. 1990. Quantitative Pt and Rh distributions
in pollution–control catalysts, Ultramicroscopy. 34: 73–80.
Maatman, R. W. 1959. How to make a more effective platinum–alumina catalyst, Ind. Eng.
Chem. 51: 913–914.
Maatman, R. W., and Prater, C. D. 1957. Adsorption and exclusion in impregnation of
porous catalytic supports, Ind. Eng. Chem. 49: 253–257.
Machek, V., Hanika, J., Sporka, K., Ruzicka, V., Kunz, J., and Janacek, L. 1983a. The
influence of solvent nature of chloroplatinic acid used for support impregnation on the
distribution, dispersity and activity of platinum hydrogenation catalysts. In Preparation
of Catalysts III, eds. Poncelet, G., Grange, P., and Jacobs, P. A., pp. 69–80. Amsterdam:
Elsevier.
Machek, V., Ruzicka, V., Sourkova, M., Kunz, J., and Janacek, L. 1983b. Preparation of
Pt/activated carbon and Pt/alumina catalysts by impregnation with platinum complexes,
Collect. Czech. Chem. Commun. 48: 517–526.
Maitra, A. M., Cant, N. W., and Trimm, D. L. 1986. The preparation of tungsten based
catalysts by impregnation of alumina pellets, Appl. Catal. 27: 9–19.
Mang, Th., Breitscheidel, B., Polanek, P., and Knoezinger, H. 1993. Adsorption of platinum
complexes on silica and alumina: Preparation of non-uniform metal distributions within
support pellets, Appl. Catal. A 106: 239–258.
Markos, J., and Brunovska, A. 1989. Optimal catalyst activity distribution in fixed-bed
reactor with catalyst deactivation, Collect. Czech. Chem. Commun. 54: 375–387.
Markos, J., Brunovska, A., and Letkova, Z. 1990. Optimal catalyst pellet activity distri-
butions. Fixed-bed reactor with catalyst deactivation, Comput. Chem. Eng. 14: 1317–
1322.
Mars, P., and Gorgels, M. J. 1964. Hydrogenation of acetylene – a theory of selectivity. In
Chemical Reaction Engineering: Proceedings of the Third European Symposium, Sup-
plement to Chem. Eng. Sci., pp. 55–65. Oxford: Pergamon Press.
References 213
Martin, G. R., White, C. W., III, and Dadyburjor, D. B. 1987. Design of zeolite/silica–alumina
catalysts for triangular cracking reactions, J. Catal. 106: 116–124.
Masi, M., Sangalli, M., Carra, S., Cao, G., and Morbidelli, M. 1988. Kinetics of ethylene
hydrogenation on supported platinum. Analysis of multiplicity and nonuniformly active
catalyst particle behavior, Chem. Eng. Sci. 43: 1849–1854.
Mathur, I., Bakhshi, N. N., and Mathews, J. F. 1972. Effect of ultrasonic waves on the activity
of chromia–alumina catalyst, Can. J. Chem. Eng. 50: 344–348.
McCoy, M. 1999. Catalyst makers look for growth, Chemical and Engineering News,
September 1999, pp. 17–25.
Melis, S., Varma, A., and Pereira, C. J. 1997. Optimal distribution of catalyst for a case
involving heterogeneous and homogeneous reactions, Chem. Eng. Sci. 52: 165–169.
Melo, F., Cervello, J., and Hermana, E. 1980a. Impregnation of porous supports – I. The-
oretical study of the impregnation of one or two active species, Chem. Eng. Sci. 35:
2165–2174.
Melo, F., Cervello, J., and Hermana, E. 1980b. Impregnation of porous supports – II. System
Ni/Ba on alumina, Chem. Eng. Sci. 35: 2175–2184.
Michalko, E. 1966a. Method for oxidizing gaseous combustible waste products, U.S. Patent
3,259,454.
Michalko, E. 1966b. Preparation of catalyst for the treatment of combustible waste products,
U.S. Patent 3,259,589.
Mieth, J. A., Schwarz, J. A., Huang, Y.-J., and Fung, S. C. 1990. The effect of chloride on
the point of zero charge of γ -Al2 O3 , J. Catal. 122: 202–205.
Minhas, S., and Carberry, J. J. 1969. On the merits of partially impregnated catalysts,
J. Catal. 14: 270–272.
Morbidelli, M., and Varma, A. 1982. Optimal catalyst activity profiles in pellets 2. The
influence of external mass transfer resistance, Ind. Eng. Chem. Fundam. 21: 284–289.
Morbidelli, M., and Varma, A. 1983. On shape normalization for non-uniformly active
catalyst pellets II, Chem. Eng. Sci. 38: 297–305.
Morbidelli, M., Servida, A., and Varma, A. 1982. Optimal catalyst activity profiles in pellets
1. The case of negligible external mass transfer resistance, Ind. Eng. Chem. Fundam. 21:
278–284.
Morbidelli, M., Servida, A., Paludetto, R., and Carra, S. 1984a. Optimal catalyst design for
ethylene oxide synthesis, J. Catal. 87: 116–125.
Morbidelli, M., Servida, A., and Varma, A. 1984b. Optimal distribution of immobilized
enzyme in a pellet for a substrate-inhibited reaction, Biotechnol. and Bioeng. 26: 1508–
1510.
Morbidelli, M., Servida, A., Carra, S., and Varma, A., 1985. Optimal catalyst activity profiles
in pellets 3. The nonisothermal case with negligible external transport limitations, Ind.
Eng. Chem. Fundam. 24: 116–119.
Morbidelli, M., Servida, A., and Varma, A. 1986a. Optimal catalyst activity profiles in
pellets. 4. Analytical evaluation of the isothermal fixed-bed reactor, Ind. Eng. Chem.
Fundam. 25: 307–313.
Morbidelli, M., Servida, A., Carra, S., and Varma, A. 1986b. Optimal catalyst activity profiles
in pellets. 5. Optimization of the isothermal fixed-bed reactor, Ind. Eng. Chem. Fundam.
25: 313–321.
Morbidelli, M., Brunovska, A., Wu, H., and Varma, A. 1991. Authors’ reply to comments
by S. Pavlou et al., Chem. Eng. Sci. 46: 3328–3329.
Moss, R. L., Pope, D., Davis, B. J., and Edwards, D. H. 1979. The structure and activity of
supported metal catalysts. VIII. Chemisorption and benzene hydrogenation on palla-
dium/silica catalysts, J. Catal. 58: 206–219.
Moyers, C. G., and Baldwin, G. W. 1997. Psychrometry, evaporative cooling and solids
drying. In Perry’s Chemical Engineers’ Handbook, 7th ed., eds. Perry, R. H., Green,
D. W., and Maloney, J. O., Chapter 12. New York: McGraw-Hill.
214 References
Mulcahy, F. M., Houalla, M., and Hercules, D. M. 1987. The effect of the isoelectric point
on the adsorption of molybdates on fluoride modified aluminas, J. Catal. 106: 210–215.
Mulcahy, F. M., Fay, M. J., Proctor, A., Houalla, M., and Hercules, D. M. 1990. The adsorption
of metal oxyanions on alumina, J. Catal. 124: 231–240.
Naoki, B., Katsuyuki, O., and Shigeki, S. 1996. Numerical approach for improving the
conversion characteristics of exhaust catalysts under warming-up condition, SAE Paper
962076.
Narsimhan, G. 1976. Catalyst dilution as a means to establish an optimum temperature
profile, Ind. Eng. Chem. Proc. Des. Dev. 15: 302–307.
Neimark, A. V., Fenelonov, V. B., and Heifets, L. I. 1976. Analysis of the drying stage in
the technology of supported catalysts, React. Kinet. Catal. Lett. 5: 67–72.
Neimark, A. V., Kheifez, L. I., and Fenelonov, V. B. 1981. Theory of preparation of supported
catalysts. Ind. Eng. Chem. Prod. Res. Dev. 20: 439–450.
Nicolescu, I. V., Parvulescu, V., Parvulescu, V., and Angelescu, E. 1990. Preparation of the
metallic supported catalysts from precursors with different immobilization capacities on
γ -Al2 O3 surface, Rev. Roum. Chim. 35: 145–159.
Nyström, M. 1978. Effectiveness factors for non-uniform catalytic activity of a spherical
pellet, Chem. Eng. Sci. 33: 379–382.
Ochoa, O., Galiasso, R., and Andreu, P. 1979. Study of some variables involved in the
preparation of impregnated catalysts for the hydrotreating of heavy oils. In Preparation
of Catalysts II, eds. Delmon, B., Grange, P., Jacobs, P., and Poncelet, G., pp. 493–506.
Oh, S. H., and Cavendish, J. C. 1982. Transients of monolithic catalytic converters: Response
to step changes in feedstream temperature as related to controlling automobile emissions,
Ind. Eng. Chem. Prod. Res. Dev. 21: 29–37.
Oh, S. H., and Cavendish, J. C. 1983. Design aspects of poison-resistant automobile mono-
lithic catalysts, Ind. Eng. Chem. Prod. Res. Dev. 22: 509–518.
Ollis, D. F. 1972. Diffusion infuences in denaturable insolubilized enzyme catalysts, Biotech-
nol. and Bioeng. 24: 871–884.
Olsbye, U., Wendelbo, R., and Akporiaye, D. 1997. Study of Pt/alumina catalysts prepara-
tion, Appl. Catal. A 152: 127–141.
Otsuka, M., Fujiyama, T., Tsukamoto, Y., Tsuiki, H., and Ueno, A. 1987. Control of
the impregnation profile of Co in an Al2 O3 sphere, Bull. Chem. Soc. Japan 60: 2881–
2885.
Ott, R. J., and Baiker, A. 1983. Impregnation of γ -alumina with copper chloride. Equi-
librium behaviour, impregnation profiles and immobilization kinetics. In Preparation of
Catalysts III, eds. Poncelet, G., Grange P., and Jacobs, P. A., pp. 685–696. Amsterdam:
Elsevier.
Papa, J., and Shah, Y. T. 1992. An approximate solution for effectiveness factor for a square
wave shell catalyst, Chem. Eng. Commun. 113: 55–61.
Papadakis, V. G., Pliangos, C. A., Yentekakis, I. V., Verykios, X. E., and Vayenas, C. G.,
1996. Development of high performance, Pd-based, three way catalysts, Catal. Today 29:
71–75.
Papageorgiou, P. 1984. Preparation of Pt/γ -Al2 O3 pellets with internal step-distribution of
catalyst, MS Thesis, University of Notre Dame.
Papageorgiou, P., Price, D. M., Gavriilidis, A., and Varma, A. 1996. Preparation of Pt/γ -
Al2 O3 pellets with internal step-distribution of catalyst: Experiments and theory, J. Catal.
158: 439–451.
Parfitt, G. D. 1976. Surface chemistry of oxides, Pure Appl. Chem. 48: 415–418.
Park, S. H., Lee, S. B., and Ryu, D. D. Y. 1981. Design of a nonuniformly distributed
biocatalyst, Biotechnol. and Bioeng. 23: 2591–2600.
Parks, G. A. 1965. The isoelectric points of solid oxides, solid hydroxides, and aqueous
hydroxo complex systems, Chem. Rev. 65: 177–198.
References 215
Pavlou, S., and Vayenas, C. G. 1990a. Optimal catalyst activity distribution in pellets for
selectivity maximization in triangular nonisothermal reaction systems: Application to
cases of light olefin epoxidation, J. Catal. 122: 389–405.
Pavlou, S., and Vayenas, C. G. 1990b. Optimal catalyst activity profile in pellets with shell-
progressive poisoning: The case of fast linear kinetics, Chem. Eng. Sci. 45: 695–703.
Pavlou, S., Vayenas, C. G., and Dassios. G. 1991. Comments on optimal catalyst activity
profiles in pellets. VIII. General nonisothermal reacting systems with arbitrary kinetics,
Chem. Eng. Sci. 46: 3327–3328.
Pereira, C. J., Kubsh, J. E., and Hegedus, L. L. 1988. Computer-aided design of catalytic
monoliths for automobile emission control, Chem. Eng. Sci. 43: 2087–2094.
Pernicone, N., and Traina, F. 1984. Commercial catalyst preparation, Appl. Ind. Catal. 3:
1–24.
Pfefferle, L. D., and Pfefferle, W. C. 1987. Catalysis in combustion, Catal. Rev. – Sci. Eng.
29: 219–267.
Pirkle, J. C., and Wachs, I. E. 1987. Activity profiling in catalytic reactors, Chem. Eng. Prog.,
August, 29–34.
Podolski, W. F., and Kim, Y. G. 1974. Modeling the water-gas shift reaction, Ind. Eng. Chem.
Proc. Des. Dev. 13: 415–421.
Polomski, R. E., and Wolf, E.E. 1978. Deactivation of a composite automobile catalyst
pellet II. Bimolecular Langmuir reaction, J. Catal. 52: 272–279.
Post, M. F. M., and Sie, S. T. 1986. Process for the preparation of hydrocarbons, Eur. Patent
Appl. 0174696.
Pranda, P., and Brunovska, A. 1993. Optimal catalyst pellet activity distributions for deac-
tivating systems: Experimental study, Chem. Eng. Sci. 48: 3423–3430.
Price, D. M., and Varma, A. 1988. Effective diffusivity measurement through an adsorbing
porous solid, AIChE Symp. Ser. 266: 88–96
Psyllos, A., and Philippopoulos, C. 1993. Performance of a monolithic catalytic converter
used in automotive emission control: The effect of longitudinal parabolic active metal
distribution, Ind. Eng. Chem. Res. 32: 1555–1559.
Radovich, J. M. 1985. Mass transfer effects in fermentations using immobilised whole cells,
Enzyme Microb. Technol. 7: 2–10.
Ray, W. H. 1981. Advanced Process Control, New York: McGraw-Hill.
Remiarova, B., Brunovska, A., and Morbidelli, M. 1993. Sulphur poisoning of nonuniformly
impregnated Pt–alumina catalysts, Chem. Eng. Sci. 48: 1227–1236.
Reuel, R. C., and Bartholomew, C. H. 1984. Effects of support and dispersion on the CO
hydrogenation activity/selectivity properties of cobalt, J. Catal. 85: 78–88.
Righetto, L., Azimonti, G., Missana, T., and Bidoglio, G. 1995. The triple layer model
revised, Coll. and Surf. A 95: 141–157.
Roth, J. F., and Reichard, T. E. 1972. Determination and effect of platinum concentration
profiles in supported catalysts, J. Res. Inst. Catal. Hokkaido Univ. 20: 85–94.
Ruckenstein, E. 1970. The effectiveness of diluted porous catalysts, A.I.Ch.E.J. 16: 151–153.
Ruckenstein, E., and Dadyburjor, D. B. 1983. Sintering and redispersion in supported metal
catalysts, Rev. Chem. Eng. 1: 251–354.
Ruckenstein, E., and Karpe, P. 1989. Control of metal distribution in supported catalysts
by pH, ionic strength and coimpregnation, Langmuir 5: 1393–1407.
Ruckenstein, E., and Malhotra, M. L. 1976. Splitting of platinum crystallites supported on
thin, nonporous alumina films, J. Catal. 41: 303–311.
Ruthven, D. M. 1984. Principles of Adsorption and Adsorption Processes, New York: Wiley.
Rutkin, D. R., and Petersen, E. E. 1979. The effect on selectivity of the macroscopic distri-
bution of the components in a dual function catalyst, Chem. Eng. Sci. 34: 109–116.
Sadhukhan, P., and Petersen, E. E. 1976. Oxidation of napthalene in packed-bed reactor
with catalyst activity profile: A design scheme for improved reactor stability and higher
product yield, A.I.Ch.E.J. 22: 808–810.
216 References
Santacesaria, E., Carra, S., and Adami, I. 1977a. Adsorption of hexachloroplatinic acid on
γ -alumina, Ind. Eng. Chem. Prod. Res. Dev. 16: 41–44.
Santacesaria, E., Gelosa, D., and Carra, S. 1977b. Basic behavior of alumina in the presence
of strong acids, Ind. Eng. Chem. Prod. Res. Dev. 16: 45–46.
Santacesaria, E., Galli, C., and Carra, S. 1977c. Kinetic aspects of the impregnation of
alumina pellets with hexachloroplatinic acid, React. Kinet. Catal. Lett. 6: 301–306.
Santhanam, N., Conforti, T. A., Spieker, W., and Regalbuto, J. R. 1994. Nature of metal
catalyst precursors adsorbed onto oxide supports, Catal. Today 21: 141–156.
Saracco, G., and Speccia, V. 1994. Catalytic inorganic membrane reactors: Present experi-
ence and future opportunities, Catal. Rev. Sci. Eng. 36: 305–384.
Sarkany, J., and Gonzalez, R. D. 1983. Effect of pretreatment on dispersion and structure of
silica- and alumina-supported Pt catalysts, Ind. Eng. Chem. Prod. Res. Dev. 22: 548–552.
Satterfield, C. N., Pelossof, A. A., and Sherwood, T. K. 1969. Mass transfer limitations in a
trickle-bed reactor, A.I.Ch.E.J. 15: 226–234.
Scelza, O. A., Castro, A. A., Ardiles, D. R., and Parera, J. M. 1986. Modeling of the im-
pregnation step to prepare supported Pt/Al2 O3 catalysts. Ind. Eng. Chem. Fundam. 25:
84–88.
Schbib, N. S., Garcia, M. A., Gigola, C. E., and Errazu, A. F. 1996. Kinetics of front-end
acetylene hydrogenation in ethylene production, Ind. Eng. Chem. Res. 35: 1496–1505.
Scheuch, S., Kamphuis, A. J., McKay, I. R., and Walls, J. R. 1996. A two stage flu-
idized/spouted bed for the granulation of catalyst powder. In Proceedings of the 1996
IChemE Research Event, pp. 997–999.
Schilling, L. B. 1994. Catalysis and Biocatalysis Technologies. Leveraging Resources and
Targeting Performance, National Institute of Standards and Technology.
Scholten, J. J. F., and Van Montfoort, A. 1962. The determination of the free-metal surface
area of palladium catalysts, J. Catal. 1: 85–92.
Schwarz, J. A. 1992. The adsorption/impregnation of catalytic precursors on pure and com-
posite oxides, Catal. Today 15: 395–405.
Schwarz, J. A., and Heise, M. S. 1990. Preparation of metal distributions within catalyst
supports IV. Multicomponent effects, J. Coll. Interf. Sci. 135: 461–467.
Schwarz, J. A., Ugbor, C. T., and Zhang, R. 1992. The adsorption/impregnation of Pd(II)
cations on alumina, silica and their composite oxides, J. Catal. 138: 38–54.
Schwarz, J. A., Contescu, C., and Contescu, A. 1995. Methods for preparation of catalytic
materials, Chem. Rev. 95: 477–510.
Seyedmonir, S. R., Plischke, J. K., Vannice, M. A., and Young, H. W. 1990. Ethylene oxida-
tion over small silver crystallites, J. Catal. 123: 534–549.
Shadman-Yazdi, F., and Petersen, E. E. 1972. Changing catalyst performance by varying
the distribution of active catalyst within porous supports, Chem. Eng. Sci. 27: 227–237.
Shah, A. M., and Regalbuto, J. R. 1994. Retardation of Pt adsorption over oxide supports
at pH extremes: Oxide dissolution or high ionic strength? Langmuir 10: 500–504.
Sheintuch, M., Lev, O., Mendelbaum, S., and David, B. 1986. Optimal feed distribution in
reactions with maximal rate, Ind. Eng. Chem. Fundam. 25: 228–233.
Shelley, S. 1997. Destroying emissions with catalysts, Chem. Eng., July, pp. 57–60.
Sherrington, D. C., and Hodge, P. 1988. Syntheses and Separations Using Functional Poly-
mers, New York: Wiley.
Shyr, Y.-S., and Ernst, W. R. 1980. Preparation of nonuniformly active catalysts, J. Catal.
63: 425–432.
Sivaraj, Ch., Contescu, Cr., and Schwarz, J. A. 1991. Effect of calcination temperature of
alumina on the adsorption/impregnation of Pd(II) compounds, J. Catal. 132: 422–431.
Smirniotis, P. G., and Ruckenstein, E. 1993. The activity of composite catalysts: Theory and
experiments for spherical and cylindrical pellets, Chem. Eng. Sci. 48: 585–593.
Smith, T. G., and Carberry, J. J. 1975. On the use of partially impregnated catalysts for yield
enhancement in non-isothermal non-adiabatic fixed bed reactors, Can. J. Chem. Eng. 53:
347–349.
References 217
Soled, S. L., Baumgartner, J. E., Reyes, S. C., and Iglesia, E. 1995. Synthesis of eggshell
cobalt catalysts by molten salt impregnation techniques. In Preparation of Catalysts VI,
eds. Poncelet, G., Martens, J., Delmon, B., Jacobs, P. A., and Grange, P., pp. 989–997.
Souza, G. L. M., Santos, A. C. B., Lovate, D. A., and Faro, A. C. 1989. Characterization of
Ni-Mo/Al2 O3 catalysts prepared by simultaneous impregnation of the active compo-
nents, Catal. Today 5: 451–461.
Spanos, N., Vordonis, L., Kordulis, C., and Lycourghiotis, A. 1990a. Molybdenum–oxo
species deposited on alumina by adsorption. I. Mechanism of the adsorption, J. Catal.
124: 301–314.
Spanos, N., Vordonis, L., Kordulis, C., Koutsoukos, P. G., and Lycourghiotis, A. 1990b.
Molybdenum–oxo species deposited on alumina by adsorption. II. Regulation of the
surface MoVI concentration by control of the protonated surface hydroxyls, J. Catal. 124:
315–323.
Spanos, N., Kordulis, C., and Lycourghiotis, A. 1991. Development of a methodology for
investigating the adsorption of species containing catalytically active ions on the surface
of industrial carriers. In Preparation of Catalysts V, eds. Poncelet, G., Jacobs, P. A., Grange,
P., and Delmon, B., pp. 175–184. Amsterdam: Elsevier.
Spanos, N., Matralis, H. K., Kordulis, C., and Lycourghiotis, A. 1992. Molybdenum–oxo
species deposited on titania by adsorption: Mechanism of the adsorption and character-
ization of the calcined species, J. Catal. 136: 432–445.
Spielbauer, D., Zeilinger, H., and Knoezinger, H. 1993. Adsorption of palladium–ammino–
aquo complexes on γ -alumina and silica, Langmuir 9: 460–466.
Sporka, K., and Hanika, J. 1992. Development of cobalt–molybdenum hydrodesulfurization
catalysts, Collect. Czech. Chem. Commun. 57: 2501–2508.
Srinivasan, R., Liu, H. C., and Weller, S. W. 1979. Sintering of shell molybdena–alumina
Stenger, H. G., Meyer, E. C., Hepburn, J. S., and Lyman, C. E. 1988. Nitric oxide reduction
using co-impregnated rhodium on an alumina Celcor honeycomb, Chem. Eng. Sci. 43:
2067–2072.
Stern, O. 1924. Zur Theorie der elektrolytischen Doppelschicht, Z. Elektrochemie 30: 508–
516.
Stiles, A. B., and Koch, T. A. 1995. Catalyst Manufacture, 2nd ed., New York: Marcel Dekker.
Subramanian, S., and Schwarz, J. A. 1991. Adsorption of chloroplatinic acid and chloroiridic
acid on composite oxides, Langmuir 7: 1436–1440.
Subramanian, S., Obrigkeit, D. D., Peters, C. R., and Chattha, M. S. 1992. Adsorption of
palladium on γ -alumina, J. Catal. 138: 400–404.
Summers, J. C., and Ausen, S. A. 1978. Catalyst impregnation: Reactions of noble metal
complexes with alumina, J. Catal. 52: 445–452.
Summers, J. C., and Hegedus, L. L. 1978. Effects of platinum and palladium impregnation
on the performance and durability of automobile exhaust oxidizing catalysts, J. Catal.
51: 185–192.
Summers, J. C., and Hegedus, L. L. 1979. Modes of catalyst deactivation in stoichiometric
automobile exhaust, Ind. Eng. Chem. Prod. Res. Dev. 18: 318–324.
Szegner, J. 1997. Effects of nonuniform catalyst distribution on inorganic membrane reactor
performance: Experiments and theory, PhD Thesis, University of Notre Dame.
Szegner, J., Yeung, K. L., and Varma, A. 1997. Effect of catalyst distribution in a membrane
reactor: Experiments and model, A.I.Ch.E.J. 43: 2059–2072.
Tanabe, K. 1970. Solid Acids and Bases, New York: Academic Press.
Taylor, K. C. 1984. Automobile catalytic converters. In Catalysis Science and Technol-
ogy, vol. 5, eds. Anderson, J. R., and Boudart, M., pp. 119–170. Berlin: Springer-
Verlag.
Thayer, A. M. 1994. Catalyst industry stresses need for partners as key to future success,
Chem. and Eng. News, July 11, pp. 19–20.
218 References
Tosa, T., Mori, T., Fuse, N., and Chibata, I. 1969. Studies on continuous enzyme reactions.
Part V. Kinetics and industrial application of aminoacylase column for continuous optical
resolution of acyl-DL-amino acids, Agr. Biol. Chem. 33: 1047–1052.
Tronci, S., Baratti, R., and Gavriilidis, A. 1999. Catalytic converter design for minimisation
of cold-start emissions, Chem. Eng. Commun., 173: 53–77.
Tsotsis, T. T., Champagnie, A. M., Minet, R. G., and Liu, P. K. T. 1993. Catalytic membrane
reactors. In Computer Aided Design of Catalysts, eds. Becker, E. R., and Pereira, C. J.,
pp. 471–552. New York: Marcel Dekker.
Uemura, Y., and Hatate, Y. 1989. Effect of nickel concentration profile on selectivity of
acetylene hydrogenation, J. Chem. Eng. Japan 22: 287–291.
Uemura, Y., Hatate, Y., and Ikari, A. 1986. Effects of post-impregnation drying conditions
on physical properties and overall reaction rate of nickel/alumina catalysts, J. Chem. Eng.
Japan 19: 560–566.
Uemura, Y., Hatate, Y., and Ikari, A. 1987. Formation of nickel concentration profile in
nickel/alumina catalyst during post-impregnation drying, J. Chem. Eng. Japan 20: 117–
123.
Underwood, R. P., and Bell, A. T. 1987. Influence of particle size on carbon monoxide
hydrogenation over silica and lanthana supported rhodium, Appl. Catal. 34: 289–310.
Uzio, D., Giroir-Fendler, A., Lieto, J., and Dalmon, J. A. 1991. Catalytic membrane reactors:
Preparation and characterization of an active Pt/alumina membrane, Key Eng. Mater. 61:
111–116.
Van den Berg, G. H., and Rijnten, H. Th. 1979. The impregnation and drying step in catalyst
manufacturing. In Preparation of Catalysts II, eds. Delmon, B., Grange, P., Jacobs, P., and
Poncelet, G., pp. 265–277, Elsevier, Amsterdam.
Van Erp, W. A., Nanne, J. M., and Post, M. F. M. 1986. Catalyst preparation, Eur. Patent
Appl. 0178008.
Van Herwijnen, T., Van Doesburg, H., and De Jong, W. A. 1973. Kinetics of the methanation
of CO and CO2 on a nickel catalyst, J. Catal. 28: 391–402.
Van Tiep, L., Bureau-Tardy, M., Bugli, G., Djega-Mariadassou, G., Che, M., and Bond, G.
C. 1986. The effect of reduction conditions on the chloride content of Ir/TiO2 catalysts
and their activity for benzene hydrogenation, J. Catal. 99: 449–460.
Varghese, P., and Wolf, E. E. 1980. Effectiveness and deactivation of a diluted catalyst
pellet, A.I.Ch.E.J. 26: 55–60.
Vayenas, C. G., and Pavlou, S. 1987a. Optimal catalyst activity distribution and generalized
effectiveness factors in pellets: Single reactions with arbitrary kinetics, Chem. Eng. Sci.
42: 2633–2645.
Vayenas, C. G., and Pavlou, S. 1987b. Optimal catalyst distribution for selectivity maximiza-
tion in pellets: Parallel and consecutive reactions, Chem. Eng. Sci. 42: 1655–1666.
Vayenas, C. G., and Pavlou, S. 1988. Optimal catalyst distribution for selectivity maxi-
mization in nonisothermal pellets: The case of parallel reactions, Chem. Eng. Sci. 43:
2729–2740.
Vayenas, C. G., and Verykios, X. E. 1989. Optimisation of catalytic activity distributions in
porous pellets. In Handbook of Heat and Mass Transfer, vol. 3, ed. Cheremisinoff, N. P.,
pp. 135–181. Houston: Gulf Publishing Company.
Vayenas, C. G., Pavlou, S., and Pappas, A. D. 1989. Optimal catalyst distribution for selec-
tivity maximization in nonisothermal pellets: The case of consecutive reactions, Chem.
Eng. Sci. 44: 133–145.
Vayenas, C. G., Verykios, X. E., Yentekakis, I. B., Papadakis, E. G., and Pliangos, C. A. 1994.
New three-way catalysts with Pt, Rh and Pd, each supported on a separate support, Eur.
Patent Appl. 0665047A1.
Vazquez, P. G., Caceres, C. V., Blanco, M. N., and Thomas, H. J. 1993. Influence of support
on concentration profiles in alumina pellets impregnated with molybdenum solutions,
Int. Commun. Heat Mass Transfer 20: 631–642.
References 219
Verykios, X. E., Kluck, R. W., and Johnson, D. L. 1983. Fixed-bed reactor simulation with
nonuniformly activated catalyst pellets. In Modeling and Simulation in Engineering, eds.
Ames, W. F., and Vichnevetsky, R., pp. 3–10. Amsterdam: North-Holland.
Vieth, W. R., Mendiratta, A. K., Mogensen, A. O., Saini, R., and Venkatasubramanian, K.
1973. Mass transfer and biochemical reaction in enzyme membrane reactor systems – I.
Single enzyme reactions, Chem. Eng. Sci. 28: 1013–1020.
Villadsen, J. 1976. The effectiveness factor for an isothermal pellet with decreasing activity
towards the pellet surface, Chem. Eng. Sci. 31: 1212–1213.
Villadsen, J., and Michelsen, M. L. 1978. Solution of Differential Equation Models by Poly-
nomial Approximation. Englewood Cliffs, NJ: Prentice-Hall.
Vincent, R. C., and Merrill, R. P. 1974. Concentration profiles in impregnation of porous
Voltz, S. E., Morgan, C. R., Liederman, D., and Jacob, S. M. 1973. Kinetic study of carbon
monoxide and propylene oxidation on platinum catalysts, Ind. Eng. Chem. Prod. Res.
Dev. 12: 294–301.
Vordonis, L., Koutsoukos, P. G., and Lycourghiotis, A. 1984. Regulation of the point of
zero charge and surface acidity constants of γ -Al2 O3 using sodium and fluoride ions as
modifiers, J. Chem. Soc. Chem. Commun. 1309–1310.
Vordonis, L., Koutsoukos, P. G., and Lycourghiotis, A. 1986a. Development of carriers with
controlled concentration of charged surface groups in aqueous solutions. I. Modification
of γ -Al2 O3 with various amounts of sodium ions, J. Catal. 98: 296–307.
Vordonis, L., Koutsoukos, P. G., and Lycourghiotis, A. 1986b. Development of carriers with
controlled concentration of charged surface groups in aqueous solutions. II. Modification
of γ -Al2 O3 with various amounts of lithium and fluoride ions, J. Catal. 101: 186–194.
Vordonis, L., Koutsoukos, P. G., and Lycourghiotis, A. 1990. Adsorption of molybdates on
doped γ -aluminas in alkaline solutions, Coll. and Surf. 50: 353–361.
Vordonis, L., Spanos, N., Koutsoukos, P. G., and Lycourghiotis, A. 1992. Mechanism of
adsorption of Co2+ and Ni2+ ions on the pure and fluorinated γ -alumina/electrolyte
solution interface, Langmuir 8: 1736–1743.
Wang, J. B., and Varma, A. 1978. Effectiveness factors for pellets with step-distribution of
catalyst, Chem. Eng. Sci. 33: 1549–1552.
Wang, L., and Hall, W. K. 1980. On the genesis of molybdena–alumina catalyst, J. Catal.
66: 251–255.
Wang, L., and Hall, W. K. 1982. The preparation and genesis of molybdena–alumina and
related catalyst systems, J. Catal. 77: 232–241.
Wang, T., and Schmidt, L. D. 1981. Intraparticle redispersion of Rh and Pt–Rh particles on
SiO2 and Al2 O3 by oxidation–reduction cycling, J. Catal. 70: 187–197.
Wanke, S. E., and Flynn, P. C. 1975. The sintering of supported metal catalysts, Catal. Rev. –
Sci. Eng. 12: 93–135.
Wei, J., and Becker, E. R. 1975. The optimum distribution of catalytic material on support
layers in automotive catalysis, Adv. Chem. Ser. 143: 116–132.
Westall, J. C. 1986. Reactions at the oxide–solution interface: Chemical and electrostatic
models, ACS Symp. Ser. 323: 54–78.
Westall, J., and Hohl, H. 1980. A comparison of electrostatic models for the oxide/solution
interface, Adv. Coll. Interf. Sci. 12: 265–294.
White, C. W., III, and Dadyburjor, D. B. 1989. Effect of nonuniform deactivation on the
optimum design of composite cracking catalysts, Chem. Eng. Commun. 86: 113–124.
Widdecke, H., Klein, J., and Haupt, U. 1986. The influence of matrix structure on the activity
and selectivity of polymer supported catalysts, Macromol. Chem., Macromol. Symp. 4:
145–155.
Wilson, G. R., and Hall, W. K. 1970. Studies of the hydrogen held by solids. XVIII. Hydrogen
and oxygen chemisorption on alumina- and zeolite-supported platinum, J. Catal. 17: 190–
206.
220 References
Wolf, E. E. 1977. Activity and lifetime of a composite automobile catalyst pellet, J. Catal.
47: 85–91.
Wu, C. H., and Hammerle, R. H. 1983. Development of low cost, thermally stable, mono-
lithic three-way catalyst system, Ind. Eng. Chem. Prod. Res. Dev. 22: 559–565.
Wu, H. 1994. Application of optimal catalyst activity distribution theory, A.I.Ch.E.J. 40:
2060–2064.
Wu, H. 1995. Optimal feed distribution in isothermal fixed-bed reactors for parallel reac-
tions, Chem. Eng. Sci. 50: 441–451.
Wu, H., Yuan, Q., and Zhu, B. 1988. An experimental investigation of optimal active catalyst
distribution in nonisothermal pellets, Ind. Eng. Chem. Res. 27: 1169–1174.
Wu, H., Brunovska, A., Morbidelli, M., and Varma, A. 1990a. Optimal catalyst activity
profiles in pellets VIII. General nonisothermal reacting systems with arbitrary kinetics,
Chem. Eng. Sci. 45: 1855–1862; 46: 3328–3329.
Wu, H., Yuan, Q., and Zhu, B. 1990b. An experimental study of optimal active catalyst
distribution in pellets for maximum selectivity, Ind. Eng. Chem. Res. 29: 1771–1776.
Xidong, H., Yongnian, Y., and Jiayu, Z. 1988. Influence of soluble aluminium on the state
and surface properties of platinum in a series of reduced platinum/alumina catalysts,
Appl. Catal. 40: 291–313.
Xue, J., Huang, Y.-J., and Schwarz, J. A. 1988. Interaction between iridium and platinum
precursors in the preparation of iridium–platinum catalysts, Appl. Catal. 42: 61–76.
Yates, D. E., Levine, S., and Healy, T. W. 1974. Site-binding model of the electrical double
layer at the oxide/water interface, J. Chem. Soc. Faraday Trans. I 70: 1807–1818.
Ye, J., and Yuan, Q. 1992. Optimal catalyst activity distribution for effectiveness factor,
productivity and selectivity maximization in generalized nonisothermal reacting systems
with arbitrary kinetics, Chem. Eng. Sci. 47: 615–621.
Yeung, K. L., Aravind, R., Zawada, R. J. X., Szegner, J., Cao, G., and Varma, A. 1994.
Nonuniform catalyst distribution for inorganic membrane reactors: Theoretical consid-
erations and preparation techniques, Chem. Eng. Sci. 49: 4823–4838.
Yeung, K. L., Aravind, R., Szegner, J., and Varma, A. 1996. Metal composite membranes:
Synthesis, characterisation and reaction studies, Stud. Surf. Sci. Catal., 101: 1349–1358.
Yeung, K. L., Sebastian, J. M., and Varma, A. 1997. Mesoporous alumina membranes:
Synthesis, characterization, thermal stability and nonuniform distribution of catalyst, J.
Membr. Sci. 131: 9–28.
Yeung, K. L., Gavriilidis, A., Varma, A., and Bhasin, M. M. 1998. Effects of 1,2-
dichloroethane addition on the optimal silver catalyst distribution in pellets for epoxi-
dation of ethylene, J. Catal. 174: 1–12.
Yortsos, Y. C., and Tsotsis, T. T. 1981. On the relationship between the effectiveness factor
for the Robin and Dirichlet problem for a catalyst with variable catalytic activity, Chem.
Eng. Sci. 36: 1734–1736.
Yortsos, Y. C., and Tsotsis, T. T. 1982a. Asymptotic behavior of the effectiveness factor for
variable activity catalysts, Chem. Eng. Sci. 37: 237–243.
Yortsos, Y. C., and Tsotsis, T. T. 1982b. On the relationship between the effectiveness
factors for the Robin and Dirichlet problems for a catalyst with nonuniform catalytic
activity. The case of generalized isothermal and nonisothermal kinetics, Chem. Eng. Sci.
37: 1436–1437.
Zaki, M. I., Mansour, S. A. A., Taha, F., and Mekhemer, G. A. H. 1992. Chromia on
silica and alumina catalysts: Surface structural consequences of interfacial events in the
impregnation course of aquated Cr(III) ions, Langmuir 8: 727–732.
Zaman, J., and Chakma, A. 1994. Inorganic membrane reactors, J. Membr. Sci., 92: 1–28.
Zhang, R., and Schwarz, J. A. 1992. Design of inhomogeneous metal distributions within
catalyst particles, Appl. Catal. A. 91: 57–65.
Zhukov, A. N. 1996. Dependences of the point of zero charge and the isoelectric point
of amphoteric solid surface on concentration and degree of binding of the ions of the
background electrolyte. The case of 1 : 1 electrolyte, Coll. J. (Engl. Ed.) 58: 270–272.
Author Index
Abello, M. C., 138 Blachou, V. M., 133, 135, 137

Aben, P. C., 44 Blanco, M. N., 170
Adcock, D. S., 107 Bolan, N. S., 149
Agashe, K. B., 135 Bonivardi, A. L., 138
Ahn, J. H., 126, 128 Bonneviot, L., 41, 144
Akratopulu, K. C., 137 Borchert, A., 122
Alexiou, M. S., 163 Bothe, N., 126
Anderson, J. R., 41 Bournonville, J. P., 42
Anderson, R. B., 134, 159, 170 Bowden, J. W., 144, 145, 147, 149
Arakawa, H., 44 Brunelle, J. P., 41–44, 136–138
Arbabi, S., 118 Brunovska, A., 23–25, 89–93
Ardiles, D. R., 2, 3, 164, 170, 173 Buchholz, K., 121, 122
Aris, R., 4, 16, 19, 23, 54, 150–156, 160, 162, Burch, R., 159
165–168 Butt, J. B., 15
Asua, J. M., 118
Au, S. S., 66 Calderbank, P. H., 83
Ausen, S. A., 159, 160 Caldwell, A. D., 83
Cannon, K. C., 102
Bacaros, T., 87 Carballo, L., 43
Baiker, A., 160, 169, 170, 172 Carberry, J. J., 2, 15, 88
Bailey, J. E., 2 Carleysmith, S. W., 122
Baldwin, G. W., 165 Castro, A. A., 133, 134, 170
Baltanas, M. A., 138 Cavendish, J. C., 114, 115
Baratti, R., 4, 30, 31, 47, 50, 53, 57–60, 62, 63, 79, Cervello, J., 4, 159, 170
81, 82, 100 Chakma, A., 95
Barbier, J., 44 Chang, H. N., 124
Barrow, N. J., 144, 145, 149 Charmas, R., 144, 149
Bartholomew, C. H., 43, 44 Chary, K. V. R., 44
Barto, M., 23, 24 Che, M., 41, 42, 144
Basset, J. M., 44, 45 Chee, Y. C., 128
Beck, D. D., 113 Chemburkar, R. M., 18, 19, 21–23, 63, 64
Becker, E. R., 2, 3, 6, 7, 69, 70, 86, 87, 162, 163, Chen, H.-C., 134, 159, 164, 167, 170, 174
168–170 Cheng, W. C., 134, 165, 170
Beeckman, J. W., 4 Chiang, C.-L., 117, 118
Bell, A. T., 44, 46 Chibata, I., 121
Benesi, H. A., 41 Chou, T. C., 93, 94, 170
Benjamin, M. M., 134, 138 Chu, P., 138, 169, 170, 177, 179
Bennett, C. O., 42 Chung, B. H., 124
Berenblyum, A. S., 67 Clause, O., 134, 138
Berger, D., 165 Contescu, C., 136–141
221
222 Author Index
Corbett, W. E., 2, 3 Hammerle, R. H., 114

Cot, L., 109 Hanika, J., 2, 134, 159, 172
Coulson, J. M., 165 Harold, M. P., 63
Cukierman, A. L., 2 Harriott, P., 79, 160, 167, 169, 170
Cusumano, J. A., 1 Hatate, Y., 67, 68
Haworth, A., 134, 144
D’Aniello, M. J., 136 Hayes, K. F., 149
D’Espinose de la Caillerie, J.-B., 135 Hayes, M. J., 159
Dadyburjor, D. B., 43, 119–121 Heck, R. M., 111
Daugherty, D. J., 2 Hegedus, L. L., 1, 3, 4, 111, 112, 162–164, 172
Davis, J. A., 147 Heise, M. S., 133–135, 140, 142, 160, 161, 163,
De Jong, K. P., 131 164, 170, 171, 177
De Miguel, S. R., 162, 164, 173 Hepburn, J. S., 94, 134, 158, 162, 163, 168,
DeLancey, G. B., 3, 89 170–172, 177
Delannay, F., 43 Hiemstra, T., 138
Delmon, B., 40, 43, 118 Hodge, P., 124
Do, D. D., 2, 123, 154 Hoekstra, J., 111, 162
Dorling, T. A., 42, 44 Hohl, H., 136, 138, 144, 149
Dougherty, R. C., 4 Hollewand, M. P., 166
Downing, D. M., 15 Holstein, W. L., 160, 169, 170, 172
Duncombe, P. R., 170 Horvath, C., 122
Hossain, M. M., 123
Ehlova, V., 2 Houalla, M., 40, 43
Engasser, J.-M., 122 Hsieh, H. P., 95
Engels, S., 134, 142 Hu, Z., 113
Ernst, W. R., 2, 133, 134, 142, 161–163, 170 Huang, C. P., 136
Huang, Y.-J., 44, 134, 138
Fang, Z.-R., 118
Farrauto, R. J., 43, 111 Iannibello, A., 134
Fenelonov, V. B., 166 Iglesia, E., 68
Figueras, F., 44 Ihm, S. K., 128
Finlayson, B. A., 55
Flynn, P. C., 43 Jianguo, W., 133, 134, 142
Foger, K., 43, 136 Jiratova, K., 137
Frost, A. C., 85 Johnson, D. L., 2
Fuentes, S., 44 Juang, H.-D., 2–4, 123
Fujiyama, T., 163, 170
Furusaki, S., 121 Karakonstantis, L., 138
Karpe, P., 140, 156, 157
Galiasso, R., 167, 170 Kasaoka, S., 2
Gates, G. C., 128 Kehoe, J. P. G., 15
Gavriilidis, A., 4, 41, 44, 46, 63, 64, 132 Kehoe, P., 2
Gazimzyanov, N. R., 170 Keil, F. J., 4
Gelosa, D., 129, 130 Keller, J. B., 35, 101, 102
Gladden, L. F., 166 Kempling, J. C., 170
Gonzalez, R. D., 43 Kheifets, L. I., 165
Gorgels, M. J., 2, 67 Kim, D. S., 138
Gottifredi, J. C., 2 Kim, Y. G., 6
Goula, M. A., 118, 136, 160, 161, 164, 167, Kimura, H., 170
169, 170 King, D. L., 44
Guisan, J. M., 122 Klein, J., 121, 124–127
Gumbleton, J. J., 112 Klugherz, P. D., 79
Ko, S. H., 170
Hachiya, K., 149 Koch, T. A., 132
Hacskaylo, J. J., 102 Komiyama, M., 3, 4, 67, 113, 134, 138, 152, 167,
Hall, W. K., 43, 134, 138 169, 170, 178
Hamada, H., 42 Kotter, M., 63, 167
Author Index 223
Kresge, C. T., 164, 170 Nuttall, T. A., 162, 163, 168–170

Kulkarni, S. S., 152, 171 Nystrom, M., 2
Kummer, J. T., 111
Kummert, R., 134 Ochoa, O., 167, 170
Kunimori, K., 83, 169, 177 Oh, S. H., 114, 115
Ollis, D. F., 124
Leckie, J. O., 134, 138, 147, 149 Olsbye, U., 135
Lee, C. K., 73–75, 77, 83–85 Otsuka, M., 163, 170
Lee, H. H., 2, 151, 152 Ott, R. J., 160, 170
Lee, J. W., 15
Lee, S. H., 43, 120 Papa, J., 2
Lee, S.-Y., 4, 150–156, 160, 162, 165–168 Papadakis, V. G., 113
LePage, J.-F., 41 Papageorgiou, P., 133, 134, 142, 143, 162–164,
Letkova, Z., 28 170, 174–178
Li, B., 162 Parfitt, G. D., 136
Li, W. D., 167, 170 Park, S. H., 122
Lilly, M. D., 121 Parks, G. A., 137
Limbach, K. W., 116–118 Pavlou, S., 18, 25, 28–30, 34, 64, 87, 88
Lin, T.-B., 93, 94, 170 Pei, D. C. T., 165
Lundquist, E. G., 129 Pereira, C. J., 1, 4, 134, 165, 170
Luss, D., 2, 3, 63 Pernicone, N., 132
Lycourghiotis, A., 137 Petersen, E. E., 2, 3, 83
Lyman, C. E., 164, 165, 170 Pfefferle, L. D., 74
Pfefferle, W. C., 74
Maatman, R. W., 159, 160, 162 Philippopoulos, C. J., 133, 135, 137
Machek, V., 134, 159, 172 Philippopoulos, C., 115
Maitra, A. M., 159, 160, 170 Pirkle, J. C., 83
Malhotra, M. L., 43 Podolski, W. F., 6
Mang, Th., 134, 135, 141, 149 Polomski, R. E., 87
Marecot, P., 44 Post, M. F. M., 68
Markos, J., 90, 91 Pranda, P., 92, 93
Mars, P., 2, 67 Prater, C. D., 159, 160
Martin, G. R., 121 Price, D. M., 174
Masi, M., 66 Psyllos, A., 115
Mathur, I., 170
McCabe, R., 3 Radovich, J. M., 121
McCoy, M., 1 Ray, W. H., 54, 183
McDowall, I. C., 107 Regalbuto, J. R., 135, 140
Melis, S., 74, 76 Reichard, T. E., 159, 170
Melo, F., 134, 150, 160, 163, 164, 170, 173 Remiarova, B., 92
Merrill, R. P., 151 Reuel, R. C., 44
Michalko, E., 2, 111, 162 Richardson, J. F., 165
Michelsen, M. L., 55, 56 Rieckmann, C., 4
Mieth, J. A., 137 Riekert, L., 63, 167
Minhas, S., 2 Righetto, L., 149
Mitchell, P. C., 134 Rijnten, H. Th., 162, 167, 170
Morbidelli, M., 2, 3, 9–16, 18, 28, 30, 34, 64, 69, Roth, J. F., 159, 170
73, 74, 123 Ruckenstein, E., 43, 119, 120, 140, 156, 157
Moss, R. L., 42, 44, 45 Ruthven, D. M., 131
Moyers, C. G., 165 Rutkin, D. R., 3
Mulcahy, F. M., 137, 138
Muraki, H., 3, 4, 113 Sadhukhan, P., 83
Sahimi, M., 118
Naoki, B., 115 Sakata, Y., 2
Narsimhan, G., 83 Santacesaria, E., 133–135, 160, 172
Neimark, A. V., 165, 166 Santhanam, N., 137, 167, 170
Nicolescu, I. V., 159, 160 Saracco, G., 95
224 Author Index
Sarkany, J., 43 Van den Berg, G. H., 162, 167, 170

Satterfield, C. N., 170 Van Erp, W. A., 68
Scelza, O. A., 134, 162, 173 Van Herwijnen, T., 6
Schbib, N. S., 6 Van Montfoort, A., 44
Scheuch, S., 131 Van Tiep, L., 41
Schilling, L. B., 1 Varghese, P., 119
Schmidt, L. D., 43 Varma, A., 2, 14, 63, 64, 75, 77, 83–85,
Scholten, J. J. F., 44 132, 174
Schwarz, J. A., 40, 44, 133–136, 140–142, 152, Vass, M. I., 136, 141
160, 161, 163, 164, 170, 171, 177 Vayenas, C. G., 4, 18, 25–30, 64, 87,
Sermon, P. A., 163 88, 113
Seyedmonir, S. R., 44 Vazquez, P. G., 154
Shadman-Yazdi, F., 2, 3 Verykios, X. E., 2–4
Shah, A. M., 135, 140 Vieth, W. R., 121
Shah, Y. T., 2 Villadsen, J., 2, 55, 56
Sheintuch, M., 78 Vincent, R. C., 151
Shelley, S., 1 Voltz, S. E., 6
Sherrington, D.C., 124 Vordonis, L., 137, 138
Shyr, Y.-S., 133, 134, 142, 161–163, 170
Sie, S. T., 68 Wachs, I. E., 83
Sivaraj, Ch., 134, 137, 138 Wang, J. B., 2
Smirniotis, P. G., 119 Wang, L., 134, 138
Smith, T. G., 2 Wang, T., 43
Soled, S. L., 159 Wanke, S. E., 43
Souza, G. L. M., 170 Wei, J., 2, 3, 6, 7, 69, 70, 86, 87, 116–118
Spanos, N., 136, 137, 138 Weller, S. W., 170
Speccia, V., 95 Weng, H.-S., 2, 3, 123
Spielbauer, D., 138, 141, 149 Westall, J., 144, 149
Sporka, K., 134 Westall, J. C., 141, 144
Srinivasan, R., 170 White, C. W., 121
Stenger, H. G., 94 Widdecke, H., 129
Stern, O., 147 Wilson, G. R., 43
Stiles, A. B., 132 Wolf, E. E., 87, 119
Stumm, W., 134, 136, 138 Wu, C. H., 114
Subramanian, S., 133–135, 138 Wu, H., 3, 30, 38, 66, 77–80, 92, 132
Summers, J. C., 111, 112, 159, 160, 162, 164
Szegner, J., 44, 45, 96, 100, 103–105 Xidong, H., 133, 135
Xue, J., 143
Tanabe, K., 134
Taylor, K. C., 111 Yates, D. E., 148
Thayer, A. M., 1 Ye, J., 39
Tiou, H.-H., 117 Yeung, K. L., 64, 65, 95–100, 103,
Tosa, T., 121 105–109, 188
Traina, F., 132 Yortsos, Y. C., 2
Tronci, S., 115 Yuan, Q., 39
Tsotsis, T. T., 2, 95
Zaki, M. I., 138
Uemura, Y., 67, 68, 167, 170 Zaman, J., 95
Underwood, R. P., 44, 46 Zhang, R., 152
Uzio, D., 105 Zhukov, A. N., 136
Subject Index
Acetylene hydrogenation, 67–68 Butadiene hydrogenation, 67

Adsorption, Butene isomerization, 128
effect of
calcination, 134 Calcination, 42–43
coimpregnants, 142–144 Carbon monoxide methanation, 66
crystallinity of support, 135 in presence of coking, 91–92
ionic strength, 135, 140 Carbon monoxide oxidation, 63–64, 69–70,
isoelectric point, 136–137 83–85, 114
pH, 135–138 in presence of poisoning, 87
precursor speciation, 140–142 Catalyst,
solvent, 144 dispersion, (see also surface area)
surface heterogeneity, 138–140 factors affecting, 40–43
finite rate, 153 rhodium, dependence on loading, 46
instantaneous, 152 layered, 111–113
nonspecific, 145 precursor,
of hexachloroplatinic acid, 133 crystallization, 41–42, 165–166
on oxides, 136–137 interaction with support, 41, 140–141
parameters for citric acid, 142 redistribution, 157–158, 166–167
parameters for hexachloroplatinic speciation, 140–142
acid, 134 three-way, 110
specific, 145 usage, 1
Adsorption isotherm, (see also surface warm-up (see lightoff)
ionization models) Catalyst distribution,
Freundlich, 134 axial, 113–115
Langmuir, 133 Dirac, 10, 71
competitive, 143 egg-white, 6
multisite model, 138 egg-yolk, 6
Air pollution, 110 eggshell, 6
Automotive exhaust catalysts, 110–115 measurement, 169
multiple step, 55, 59–63
Bang-bang control system, 54–55 nickel, 179
Benzene nonuniform, 1–4
alkylation, 128 platinum, 104, 161, 175
hydrogenation, 66 rhodium, 168, 172
Benzoyl-L-arginine ethyl ester hydrolysis, 122 step, 12–13, 55–57, 101
Benzylpenicillin sulfonic group, 127
deacylation, 122 Catalyst pellet,
hydrolysis, 122 basic equations,
Biocatalysts, 121–124 multiple reactions, 35–36, 47–48
Biot number, 14, 19 single reaction, 7–8, 15–16, 31, 51
225
226 Subject Index
Catalytic combustion, 74 Isoprene hydrogenation, in presence of coking,

Catalytic cracking, 119–121 93–94
Cold-start emissions (see lightoff)
Conversion, optimal, 74 Jacobian, 49
Cyclohexane dehydrogenation, 102–103
Kinetics,
Deactivation (see also poisoning) Eley-Rideal, 67
effect on pellet dynamics, 24–25 Langmuir-Hinshelwood, 6, 9, 16, 23, 60
enzyme, 123–124 Michaelis-Menten, 121, 123
Diffusivity, evaluation, 174
Dienes hydrogenation, 67 Lagrange multipliers, 32–33, 49–50, 52, 61,
Drying, 102, 183
effect on catalyst dispersion, 41–42 Lightoff, 114–115
effect on catalyst distribution, 156–158, Local OCD approach, 71, 78–80
165–169
regimes of, 165–166 Maximum principle, 52
Dynamic behavior of pellets, 23–25 Membrane reactor,
catalytic, 96
Effectiveness factor, optimal, 10–11, 14, 17, inert, with catalyst on the feed side, 96
53–54, 57 basic equations, 188–191
Electrokinetic phenomena, 156–157 preparation, 105–109
Enzymes (see biocatalysts) with nonuniform catalyst distribution, 95–100
Ethane dehydrogenation, 103–105 Monolith, 69–70
Ethanol dehydration, in presence of two-zone, 113–114
deactivation, 128 Moving-bed reactor, 118
Ether cleavage, 129 Multiplicity,
Ethylene epoxidation, 30, 63–65, 79–83 in catalyst pellets, 23–25, 63–64
Ethylene hydrogenation, 66 in reactors, 74, 77, 83–84
in presence of poisoning, 92–93
Nitrogen oxide reduction, in presence of
External transport resistances,
poisoning, 94
effect on optimal catalyst distribution, 14–15,
Numerical optimization, 30, 55–56
18–22
Optimal active element distribution,
Fischer-Tropsch synthesis, 68 multiple reactions, 59, 62
Fixed bed (see plug-flow reactor) single reaction, 53–54, 56–58
Optimal catalyst distribution, (see also optimal
Global OCD approach, 78–80 catalyst location, optimal active element
distribution)
Hamiltonian, 49, 51, 54, 61, 186 for nonselective poisoning, 88
development of, 48–51, 192–193 in membrane reactors, 100–102
Helmholz plane, 147 Optimal catalyst location, along reactor,
Hydrodemetallation, 116–118 ethylene epoxidation, 82
Hydrodesulfurization, 118 multiple reactions, 80
single reaction, 72–74
Imbibition, 149 Optimal catalyst location, in single pellet,
Impregnation, 40–41 multiple reactions, 26–29, 39–40
dry, 150–153, 169–171 single reaction, 10–12, 14, 17–18, 19–22
basic equations, 150–151 Optimal feed distribution, 78–80
multicomponent, 152–153, 155–156, 161–165, Optimality conditions, 32–33, 38, 102
171–179 Optimization, catalyst activity distribution,
one-component, 151–152, 154, 159–161, in membrane reactors, 100–102
169–172 in plug-flow reactor,
sequential, 164, 174–178 for Langmuir-Hinshelwood kinetics, 69–74,
wet, 153–156, 171–179 77–80
basic equations, 154 homogeneous-heterogeneous reactions,
Ionic dissociation phenomena, 156–157 74–75
Ionic strength, 140 in presence of poisoning, 90–91
Isoelectric point (see point of zero multiple reactions, 77–83
charge) single reaction, 69–77
Subject Index 227
in single pellet, temperature, 161

for arbitrary kinetics, 18–22, 35–40 viscosity, 167
for first-order kinetics, 18 volume of solution, 160
for Langmuir-Hinshelwood kinetics, 8–11, multimetallic, 164
14, 16–18, 22, 26–27 Propane oxidation, 83
for power-law kinetics, 20–22, 26 Propene oxidation, 63
in presence of poisoning, 86–91 Proton affinity distribution, 139
multiple reactions, 25–30, 35–40 Proton binding isotherm, 139
single reaction, 6–23
Optimization, catalyst loading distribution, Reduction, 43
in single pellet, Resins, functionalized polymer, 124–130
for first-order kinetics, 51–54, 56–59 gelular, 125
for Langmuir-Hinshelwood kinetics, 60–63 macroreticular, 125
multiple reactions, 46–51, 58–63 preparation, 124–127
single reaction, 51–54, 56–58 Runaway, 81–83
Optimization indices (see performance indices)
Selectivity, optimal, 26–29
Performance indices, Sensitivity, of catalyst performance, 13–14, 22–23
conversion, 71, 96, 191 Slip-casting, 104–109
conversion over catalyst lifetime, 88 sequential, 105–106
effectiveness factor, 8, 36, 48, 60 Stability, 23–25
purity, 99, 192 Surface area, dependence on loading,
selectivity, 36, 48, 78, 192 catalytic, 43–47
yield, 36, 48, 78 palladium, 45
Phthalic anhydride esterification, 129–130 platinum, 44–45, 107
Plug-flow reactor, heterogeneous, silver, 46
basic equations, Surface complexation models (see surface
multiple reactions, 77–78, 80 ionization models)
single reaction, 70–71, 75–77 Surface ionization models, 146
with multiple zones, 73–74, 82–85 constant capacitance, 145–146
Point of zero charge, 136–137 diffuse layer, 146–147
Poisoning, four-layer, 149
in automotive exhaust catalysts, 111–113 Stern, 147
nonselective, 86–89 triple layer, 147–149
pore-mouth, 86–87, 119 Switching function, 54, 61
selective, 89–91
uniform, 86–87 Thiele modulus,
Preparation, nonuniform catalyst, multiple reactions, 36
by coating, 132 single reaction, 8, 10, 16
by deposition-precipitation, 131 poison, 86
by extrusion, 132
by granulation, 131 Uniform distribution, comparison with
by impregnation, 159–165, 169–179 Dirac distribution, in membrane reactors,
by tableting, 132 97–100
effect of optimal distribution,
adsorption strength, 167 hydrodemetallation, 117
coimpregnants, 162–163 multiple reactions, 26–29, 65, 68
concentration, 160, 163, 174, 177 single reaction, 57–58
doping of support, 161 step distribution, 7, 70
drying, 156–158, 165–169
electrokinetic phenomena, 156 Variational method, 48, 183
impregnation time, 160, 163, 175 Volatile organic compounds incineration, 85
impregnation type, 159, 163 Voorhies equation, 88
ionic dissociation phenomena, 156
ionic strength, 160 Yield, optimal,
nature of precursor, 159 ethylene epoxidation, 82
pH, 160 multiple reactions, 79–80
solvent, 159, 163
support surface area, 165 Zeolites, 119–121

Book Catalyst Design Optimal Distribution of Catalyst in Pellets Reactors and Membranes

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Book Catalyst Design Optimal Distribution of Catalyst in Pellets Reactors and Membranes

Uploaded by

Copyright:

Available Formats

Catalyst Design

Heterogeneous catalysis is widely used in chemical, reﬁnery, and pollution-

Massimo Morbidelli is Professor of Chemical Reaction Engineering in

Asterios Gavriilidis is Senior Lecturer in the Department of Chemical

Arvind Varma is the Arthur J. Schmitt Professor in the Department of

Books in the Series:

CAMBRIDGE UNIVERSITY PRESS

Cambridge University Press

© Cambridge University Press 2001

This publication is in copyright. Subject to statutory exception

First published 2001

Library of Congress Cataloguing in Publication data

ISBN-13 978-0-521-66059-4 hardback

ISBN-13 978-0-521-01985-9 paperback

Preface page xiii

3 Optimization of the Catalyst Distribution in a Reactor 69

7.1.3.b Diffuse-Layer Model 146

Appendix A: Application of the Maximum Principle for Optimization

Heterogeneous catalysis is used widely in chemical, reﬁnery and pollution-control

1.1 Importance of Catalysis

1.2 Nonuniform Catalyst Distributions

to catalyst performance. The effects of deliberate nonuniform distribution of the

1.3 Overview of Book Contents

2.1 The Case of a Single Reaction

2.1.1 Isothermal Conditions

Figure 2.1. Isothermal effectiveness factor η as a function of

Deﬁnition of optimization problem

the steady-state mass balance for a single pellet is given by

where De is the effective diffusivity, x is the space coordinate, C is the reactant

a(x) = k(x)/k̄ (2.3)

The boundary conditions (BCs) are

u = C/Cf , s = x/R, φ 2 = r (Cf )R2 /De Cf .

which, using equation (2.8), yields

Shape of optimal catalyst distribution

r (C) = k̄C/(1 + KC)2 (2.12)

Figure 2.2. Shape of the dimensionless

which, using equation (2.8), gives

This function exists and is given by

which physically corresponds to a sharp peak located at s = s̄. In our optimiza-

Evaluation of optimal catalyst location

Table 2.1. Location of optimal catalyst distribution and

The corresponding optimal effectiveness factor is

Figure 2.3. Optimal catalyst loca-

Step catalyst distribution

where s1 = s̄ − , s2 = s̄ + , and a is a constant which is evaluated using con-

Figure 2.4. Effectiveness factor η vs

Figure 2.5. Maximum value of effectiveness

corresponding Dirac-delta distribution, i.e., as → 0, ηm → ηopt = 5.51 as given

Sensitivity of catalyst performance to the step location

Figure 2.6. Values of the effectiveness factor η for a

Table 2.2. Location of optimal catalyst distribution and corresponding

σ≤1 Bim > Bi∗ Bim < Bi∗

Effect of external mass transfer resistance

2.1.2 Nonisothermal Conditions

Deﬁnition of optimization problem

where λe is the effective thermal conductivity, T is the temperature, and − H is

a(x) = k(x, Tf )/k̄(Tf ). (2.37)

It is thus independent of temperature. In dimensionless form, the above equations

Shape of optimal catalyst distribution

Table 2.3. Location of optimal catalyst distribution and

Evaluation of optimal catalyst location

Figure 2.7. Optimal catalyst location s̄opt as a function of

Note that σ = 0 corresponds to a ﬁrst-order reaction. In this case, under isother-

2.1.3 Arbitrary Kinetics with External Transport Resistances

where λe is the effective thermal conductivity, T is the temperature, and −H is